A kind of fast robust automatic division method of abdominal CT sequence image liver
Technical field
The invention belongs to technical field of image processing, it is related to the segmentation of abdominal CT sequence image organ, more particularly to belly
The fast robust of CT sequence image livers is split automatically, available for medical image auxiliary diagnosis and treatment.
Background technology
Liver segmentation is to realize the premise of liver diseases computer-aided diagnosis and hepatic transplantation preplanning.Utilize segmentation
And the hepatic model that reconstruction is obtained can be with Auxiliary Liver lesion analysis, cubing, vessels analysis, liver subsection, medical diagnosis on disease
With assess etc. work.CT angiographic images are small to injury of human due to high resolution, can image, reflect liver exactly
And its lesion locations and by the generally favor of doctor.Very big (the every patient of image slice quantity used due to three-dimensional imaging
Hepatic CT sequence be about 200 or so), artificial segmentation very time-consuming and segmentation result of each cutting into slices has very big subjectivity.Cause
This, studies Clinics and Practices of the fast robust automatic division method to liver diseases of liver in abdominal CT sequence image, to carrying
The precision of high computer-aided diagnosis and efficiency are significant.
Because liver organ is complicated, in irregular shape, Different Individual differs greatly, and imaging when by noise,
The influence of skew and histokinesis etc., the features such as obtained Hepatic CT sequence image generally has complexity and diversity.In addition,
The organ such as liver and the abdominal muscles of surrounding, stomach, diaphram, spleen, kidney and heart lacks preferable intensity contrast in CT images,
Accurate automatic segmentation band to liver in CT sequence images is all carried out very big difficulty by this.
Existing Hepatic CT sequences segmentation method generally can be divided into based on image and the major class of statistical model two.It is based purely on figure
The dividing method of picture refers to the method directly split with brightness, texture and other image self informations, mainly including threshold
Value method, cluster, region growing, movable contour model and figure are cut, and these cutting techniques are primarily present following shortcoming and defect:
(1) need to carry out complicated pretreatment, including remove surrounding tissue or the organs such as rib, vertebra, kidney;(2) it is difficult to segmentation pair
The fuzzy CT sequence images of, liver boundary lower than degree.Method based on statistical model uses substantial amounts of CT sequence images structure first
Build target prior model, be then applied to the segmentation of current sequence, such method for the relatively low image of contrast have compared with
Good segmentation effect, but it is poor for liver segmentation effect in irregular shape, and time-consuming, quick to data initialization and registration
Sense.
The content of the invention
The present invention has taken into full account the shortcoming and deficiency of above-mentioned prior art, and its object is to there is provided a kind of accurate, fast
Speed, the abdominal CT sequence image liver automatic division method of robust.And the abdominal CT sequence image liver segmentation side of the present invention
Method can be generalized in the segmentation of other abdomen organs.
The abdominal CT sequence image liver robust automatic division method of the present invention, comprises the following steps:
According to the brightness Probability Characteristics of liver area, the Intensity model based on gaussian curve approximation is set up, suppresses multiple
Miscellaneous background, prominent liver area;
Using image local information, the liver display model based on PCA is set up, liver weak boundary, increasing is further enhanced
The discrimination of liver and background;
Using the spatial coherence of contiguous slices, with reference to brightness and display model, structure figure cuts energy function, realizes liver
The fast robust in region is split automatically.
The Intensity model, which is set up, to be included:
A. the manual selected part liver area from list entries, using its brightness probability distribution of Gaussian function fitting, and
Substantially brightness range [the I of the sequence image liver is obtained according to the confidential interval of Gaussian Profilemin,Imax]。
B. it is directed to liver brightness range [Imin,Imax], build following Intensity model:
Wherein, k is 0.5~5 normal number, and I is brightness of image,I minWithI maxThe liver brightness that respectively step a is obtained
Minimum value and maximum.I is closer to brightness range [Imin,Imax] central value, Intensity model fintensity(I) value is bigger,
Show the pixel belong to liver probability it is bigger.
The display model, which is set up, to be included:
The subgraph that size is (2n+1) × (2n+1) is taken centered on each pixel, and is believed with the brightness of the subgraph
Breath represents the feature of the central pixel point, and wherein n is 1~8 natural number;
Principal component analysis is carried out with the PCA partial liver region all pixels neighborhood of a point subgraphs specified to user simultaneously
The average of its top n principal component is calculated, the external appearance characteristic p of liver is obtainedPCA_average, wherein N is 1~(2n+1)×(2n+ 1)
Natural number;
Using distance function as outward appearance similarity measurement, the outward appearance for calculating all pixels point and liver in whole sequence is special
Sex differernce:
D (p)=| | pPCA(p)-pPCA_average||
Wherein, pPCA(p) represent to carry out pixel p neighborhood subgraph the top n principal component that PCA feature extractions are obtained,
D (p) represents principal component feature pPCAAnd p (p)PCA_averageThe distance between;
Build the liver display model based on PCA:
Wherein, mean_dpRepresent to calculate the obtained average apart from d (p) to all pixels point p.fPCA(p) value model
Enclose for 0~1, fPCA(p) probability that bigger expression pixel p belongs to liver is bigger.
The present invention abdominal CT sequence image liver robust automatic division method be further characterized in that, the liver area it is fast
Automatically segmentation includes fast robust:
User specifies initial slice.Liver area is met in sequence relatively large and only comprising liver connected region
Sectioning image can be designated as initial slice, acquiescence be located at list entries from top to bottom 1/3rd at, if for example, input
CT sequences sum is 210, then initial slice is defaulted as the 70th, and if the slice numbers of CT sequences can not be divided exactly by 3, it is right
Division result carries out round.
Initial slice is split.First energy function is cut with the brightness and display model structure figure of the section:
Wherein, P represents all pixels collection in image f;NpRepresent pixel p neighborhood territory pixel collection;Fintensity(fp) and
FPCA(fp) it is respectively brightness penalty term and outward appearance penalty term, calculated obtain by brightness and display model respectively;α and β are respectively control
Brightness processed and the weight of outward appearance penalty term, span is 0~1, and meets alpha+beta=1;B(fp,fq) it is border penalty term, by
Gradient calculation between adjacent pixel is obtained, for controlling the smoothness for splitting curve.Then minimized using optimization algorithm
The energy function, obtains segmentation result.Its largest connected region is finally taken finally to split as the liver of the sequence initial slice
As a result.
Sequence section is split.The present invention uses the mode of iteration to split sequence up and down respectively by starting point of initial slice
All sections of other in row.Because the size and location of liver between contiguous slices is not significantly altered, therefore, in iterative segmentation
During, the liver position information between contiguous slices is also adopted as the accuracy of energy penalty term increase segmentation.Now, energy
Flow function is represented by:
Wherein, FlocationRepresent to calculate obtained position penalty term by upper a piece of segmentation result.The present invention is by using phase
Neighbour section liver position information, can effectively remove with the incoherent analogous tissue of liver, such as spleen, pancreas, muscle are obtained
To more accurately segmentation result.
Compared with prior art, the inventive method has following advantage:
The invention provides a kind of adaptive liver sequences dividing method, the data characteristics for each patient is set up
Spatial coherence between corresponding brightness and display model and combination contiguous slices is split automatically, can effectively handle shape
The different liver of shape, brightness, with stronger robustness;
Complicated preprocessing process is not needed, it is not necessary to which muscle, rib, vertebra, kidney etc. are split in advance;
Cumbersome training need not be carried out to build with statistical model, splitting speed is very fast;
The abdominal CT sequence image liver segmentation method of the present invention can be generalized to abdominal CT sequence image other organs
In segmentation application, such as segmentation of spleen and kidney.
Brief description of the drawings
The abdominal CT sequence image liver robust automatic division method flow chart of Fig. 1 embodiment of the present invention;
The partial liver area schematic that Fig. 2 embodiment of the present invention users specify;
The user of Fig. 3 embodiment of the present invention specifies liver area gaussian curve approximation result figure;
The liver Intensity model exemplary plot of Fig. 4 embodiment of the present invention;
The liver display model exemplary plot of Fig. 5 embodiment of the present invention;
The liver segmentation results exemplary plot of Fig. 6 embodiment of the present invention.
Embodiment
Embodiment 1
Fig. 1 show the abdominal CT sequence image liver robust automatic division method flow chart of embodiment of the present invention.It is first
First selected part liver area any from the CT sequences of input sets up liver Intensity model and display model respectively, then in conjunction with
Brightness and display model cut algorithm with figure the liver in initial slice are split, and finally use the mode of iteration with initial
The liver of other all sections in upward, downward sequence of partitions is distinguished in segmentation section for starting point., will be upper during iterative segmentation
The figure that the liver position information of a piece of segmentation result incorporates current slice cuts energy function, to increase the accuracy of segmentation result.
Until all section segmentations are completed, program end of run exports segmentation result.
With reference to Fig. 1, the abdominal CT sequence image liver robust of the present invention is described in detail certainly with a preferred embodiment
Dynamic dividing method.
1. liver Intensity model is set up.Implement step as follows:
According to CT scan characteristic, the brightness of some certain organs is typically located substantially in a narrower interval range, and
It is in Gaussian Profile on the interval.If liver brightness range is [I in abdominal CT sequencemin,Imax], then liver Intensity model can be built
It is as follows:
Wherein, k is 0.5~5 normal number, and the present embodiment is preferably that 1.5, I is brightness of image.I is closer to brightness range
[Imin,Imax] central value, Intensity model fintensity(I) value is bigger, shows that the pixel belongs to the probability of liver and got over
Greatly.
In order to obtain brightness range [Imin,Imax], the Luminance Distribution in the partial liver region that the present invention is specified to user is entered
Row Gauss curve fitting:
Wherein, a represents the center of Gaussian Profile, and b controls the width of Gaussian Profile, and c is the peak value of Gaussian Profile.Figure
2 be any sectioning image in original CT sequence, and wherein rectangular area represents the partial liver region that user specifies.Shown in Fig. 3
For the result of gaussian curve approximation is carried out to the liver area using least-squares algorithm, it can be seen that the brightness of liver area is general
Rate can meet Gaussian Profile well.According to the probability theory of Gaussian Profile, [a-b, a+b], [a-2b, a+2b] and [a-3b, a+
3b] brightness range can cover the pixel of liver area 68%, 95%, 99% respectively.Due to the abutting tissue compole of liver
It there may be brightness overlapping, therefore, in order at utmost suppress complex background, the preferred liver brightness range of this programme is [a-b, a
+b].Fig. 4 is the Intensity model result obtained using the present embodiment, wherein original image of first behavior from different CT sequences,
Second behavior uses the liver Intensity model that the present embodiment method is obtained.
2. liver display model is set up.Implement step as follows:
(1) 9 × 9 subgraph is taken centered on each pixel, and the center is represented with the monochrome information of the subgraph
The feature of pixel;
(2) principal component point is carried out with the PCA partial liver region all pixels neighborhood of a point subgraphs specified to user
The average of its preceding 6 principal component is analysed and calculated, the characteristic vector p of one 1 × 6 dimension is thus obtainedPCA_average, i.e. the outward appearance of liver
Feature;
(3) using Euclidean distance as outward appearance similarity measurement, the outer of all pixels point and liver in whole sequence is calculated
See feature difference:
Wherein, pPCA(p) represent to carry out pixel p subgraph preceding 6 features that PCA feature extractions are obtained, d (p) tables
Show principal component feature pPCAAnd p (p)PCA_averageBetween Euclidean distance;
(4) the liver display model based on PCA is built:
Wherein, mean_dpRepresent to calculate the obtained average apart from d (p) to all pixels point p.fPCA(p) value model
Enclose for 0~1, fPCA(p) probability that bigger expression pixel p belongs to liver is bigger.
Fig. 5 is the display model result obtained using the present embodiment, wherein the first behavior is from the original of different CT sequences
Image, the second behavior uses the liver display model that the present embodiment method is obtained.
3. initial slice is split.Implement step as follows:
After obtaining liver brightness and display model using above-mentioned specific implementation method, the figure for building initial slice cuts energy
Function:
Wherein, P represents all pixels collection in image f;NpRepresent pixel p neighborhood territory pixel collection;α and β are respectively control
Penalty term F processedintensity(fp) and FPCA(fp) weight, meet alpha+beta=1, preferred α=0.8, β=0.2 of the present embodiment;
Fintensity(fp), FPCA(fp) and B (fp,fq) it is respectively brightness, outward appearance and border penalty term, and be defined respectively as:
Wherein
IpFor the brightness value of pixel p, d (p, q) is the Euclidean distance between pixel p and q, NumPFor pixel in set of pixels P
Number.Then using the max-flow min-cut algorithmic minimizing energy function, segmentation result is obtained.Finally take its largest connected
Region as the sequence initial slice the final segmentation result of liver.
4. sequence section is split.Implement step as follows:
After above-mentioned specific implementation method segmentation initial slice, then with the mode of iteration using initial segmentation section as
Put other all sections distinguished in sequence of partitions up and down.During the iterative segmentation, split using upper one section
As a result the accuracy of segmentation is increased as liver position penalty term.Now, figure cuts energy function and is represented by:
Wherein
μ is the normal number between 0~0.5, and the present embodiment preferably 0.1, Dis (p) is the distance of upper one section segmentation result
Conversion.Fig. 6 is the part CT image liver segmentation results obtained using the present embodiment, wherein the first behavior comes from different CT sequences
Original image, the second behavior uses the obtained liver segmentation results of the present embodiment method.
Embodiment 2
The 10 Hepatic CT sequences provided using the method for embodiment 1 XHCSU14 databases are tested, and use five
Individual error criterion is evaluated test result, including:Volume aliasing error (Volumetric Overlap Error, VOE),
Opposite bank product moment (RelativeVolumeDifference, RVD), even symmetrical surface distance (Average Symmetric
SurfaceDistance, ASD), root mean square symmetrical surface distance (Root Mean Square Symmetric
SurfaceDistance, RMSD), and maximum symmetrical surface distance (Maximum Symmetric SurfaceDistance,
MSD)。
10 cycle tests of XHCSU14 databases derive from the row's multi-layer spiral CTs of Philipsbrilliance 64
Machine, is provided by Xiangya Hospital, Central-South China Univ., and slice plane number of pixels is 512 × 512, and planar pixel spacing range is 0.53-
0.74mm, interlamellar spacing is 1.0mm.XHCSU14 databases are divided using five error criterions of VOE, RVD, ASD, RMSD and MSD
Cut result to be evaluated, obtained result is as shown in table 1.It can be seen that, for the CT sequences of 10 different patients, the present invention point
The average and standard deviation for cutting error are smaller, show the liver area in the inventive method energy accurate and effective Ground Split abdominal CT sequence
Domain, and with stronger robustness.
Table 1
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
God is with principle, and any modification, equivalent substitution and improvements done etc. should be included within the scope of protection of the invention.