CN108876769A - A kind of left auricle of heart CT image partition method - Google Patents
A kind of left auricle of heart CT image partition method Download PDFInfo
- Publication number
- CN108876769A CN108876769A CN201810553854.0A CN201810553854A CN108876769A CN 108876769 A CN108876769 A CN 108876769A CN 201810553854 A CN201810553854 A CN 201810553854A CN 108876769 A CN108876769 A CN 108876769A
- Authority
- CN
- China
- Prior art keywords
- image
- sequence
- heart
- left auricle
- indicate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20156—Automatic seed setting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30048—Heart; Cardiac
Abstract
A kind of left auricle of heart CT image partition method, it is related to Medical Image Processing, noise reduction process is carried out to left auricle of heart CT image first, using the boundary information of priori knowledge building left auricle of heart missing, using area growth algorithm obtains individual initial segmentation image as individual label figure of Atlas；The Atlas segmentation that individual is marked and the information relationship of left auricle of heart CT flanking sequence improves is recycled, initial segmentation sequence is obtained；Using the profile information in initial segmentation sequence as the initial profile of level set algorithm, and the approximate Euclidean distance information for obtaining left auricle of heart CT sequence chart is inputted as feature, accurate segmentation result sequence is obtained finally by quick level set algorithm, to achieve the purpose that left auricle of heart CT image segmentation.
Description
Technical field
The present invention relates to Medical Image Processings, more particularly, to a kind of left auricle of heart CT image partition method.
Background technique
Left auricle of heart is the component part in human body atrium sinistrum, and function is auxiliary pump blood.The multiple-limb knot of left auricle of heart itself
Structure tends to interfere with blood backflow, when left auricle of heart just cannot produce contraction, atrial pressure is caused to increase, and causes atrial fibrillation, left auricle of heart knot
Structure has a major impact cardiovascular and cerebrovascular disease.When carrying out blocking operation to left auricle of heart, the shape for how being directed to left auricle of heart selects to close
Suitable plugging device is the basis of successful surgery.Therefore, left auricle of heart CT image segmentation is to carry out left auricle of heart image analysis, processing
One of basis is of great significance.However because left auricle of heart with atrium sinistrum to be adhered part non-boundary, left auricle of heart CT image texture-free
The features such as feature, left auricle of heart CT image segmentation have certain difficulty.The research of existing left auricle of heart image segmentation is also fewer, with
Maximally related the method for the present invention is that (yellow contain Cape jasmine, Han Luyi, Liu Qi, text of an annotated book Juan, Deng Lihua, what insults the left heart of ultrasound to yellow contain Cape jasmine et al.
The profile of ear image automatically extracts [J] Sichuan University journal (engineering science version), 2016,48 (03):87-93.) one proposed
The partitioning algorithm of kind left auricle of heart ultrasound image, the algorithm realize the dividing function of left auricle of heart ultrasound image, from individual ultrasound figure
Left auricle of heart Accurate Segmentation is realized as in, which first passes through the position identification of two steps, then by close to the initial of true edge
Profile obtains precisely segmentation using movable contour model, and the position determination of first two steps needs to rely on priori knowledge, algorithm robustness
It is low.And this method carries out left auricle of heart image dividing processing for ultrasound image, can not be directly used in left auricle of heart CT image segmentation.
Left auricle of heart changeable, complicated anatomical features with shape.Left auricle of heart CT image has following features：It is left
Auricle and atrium sinistrum are adhered substantially texture-free feature of part non-boundary, left auricle of heart etc., therefore the boundary of left auricle of heart CT image can not
It directly obtains, it is bad using traditional image segmentation algorithm effect.
Summary of the invention
It is an object of the invention to In view of the above shortcomings of the prior art, provide a kind of image segmentation side left auricle of heart CT
Method.
The present invention carries out noise reduction process to left auricle of heart CT image first, utilizes the boundary of priori knowledge building left auricle of heart missing
Information, using area growth algorithm obtain individual initial segmentation image as individual label figure of Atlas；Recycle individual label with
And the Atlas segmentation that the information relationship of left auricle of heart CT flanking sequence improves, obtain initial segmentation sequence；By initial segmentation sequence
Initial profile of the profile information as level set algorithm in column, and obtain the approximate Euclidean distance information of left auricle of heart CT sequence chart
It is inputted as feature, accurate segmentation result sequence is obtained finally by quick level set algorithm, to reach left auricle of heart CT image
The purpose of segmentation.
The present invention includes the following steps：
1) left auricle of heart CT sequence image A is inputted；
2) noise reduction process is carried out to the sequence image A in step 1), obtains sequence image B；
3) selecting step 2) in sequence image B in individual CT scheme Bi, residue sequence image is denoted as B '；
4) individual CT according to obtained in step 3) schemes BiThe position feature information of middle left auricle of heart manually chooses two features
Point, the boundary information of building left auricle of heart missing, obtains image C；
5) to image C using area growth algorithm obtained in step 4), image L is obtainedi；
6) the image L obtained with step 5)iFor initial markup image, the image B ' Atlas for carrying out adjacent registration is divided,
Obtain initial segmentation sequence L；
7) to the sequence image B in step 2), all foreground points is calculated to the approximate Euclidean distance of background dot, obtain distance
Image sequence D；
8) profile for the initial segmentation sequence L for obtaining step 6) is as initial profile, by distance obtained in step 7)
Image sequence D carries out the accurate of left auricle of heart CT sequence image to sequence image B as characteristic image, using quick level set algorithm
Segmentation.
In step 2), the sequence image A carries out noise reduction process, and the specific method for obtaining sequence image B can be：
(1) image of i-th image after t iterative filtering is denoted as I in sequence image At,i, use curvature anisotropy
Diffusion equation recalculates：
Wherein, c (| x |) indicates specific conductance,▽ expression takes gradient, and t indicates the number of iterations, I (x, y,
I) i-th image in sequence image A is indicated；
(2) the process n times of iterative step (1), i-th image after obtaining noise reduction；
(3) it is incremented by i, enables i=i+1, execute step (1), obtains the image sequence B after retaining edge noise reduction；
In step 3), the selecting step 2) in sequence image B in individual CT scheme BiSpecific method can be：
(1) the image number n in left auricle of heart CT sequence image A is calculated；
(2) i-th image is chosen as figure Bi, wherein i is the maximum integer no more than n/2.
In step 4), described individual CT according to obtained in step 3) schemes BiThe position feature information of middle left auricle of heart, people
Work chooses two characteristic points, the boundary information of building left auricle of heart missing, and the specific method for obtaining image C can be：
(1) two seed points manually chosen are set as p1(x1, y1)、p2(x2, y2), this two o'clock can determine a logical coordinates
Straight line in systemIn physical coordinates system,It can be by from point p1To point p2The pixel fitting extended one by one is constituted, then point p1's
Next pixel position is expressed as：
Wherein, N0Indicate the position of next pixel, N1, N2Indicate position candidate, d1, d2Respectively indicate N1, N2To logic
The distance of next point on coordinate；
(2) according to the positional relationship straight line in left auricle of heart anatomical structureSlope k meet：K ∈ (- 45 °, 0 °) ∪
(- 90 °, -45 °), then its corresponding N1, N2It is expressed as：
(3) when the slope k of straight line meets k ∈ (- 45 °, 0 °)：
Wherein, pi+1Indicate the selection criteria of the next position, dx indicates horizontal departure value in logic, and dy is indicated in logic
Vertical missing value, according to pi+1Value can determine the result of next coordinate：
Work as straight lineSlope k when meeting k ∈ (- 90 °, -45 °)：
Wherein, pi+1Indicate the selection criteria of the next position, dx indicates horizontal departure value in logic, and dy is indicated in logic
Vertical missing value.According to pi+1Value can determine the result of next coordinate：
(4) image B is modifiediMiddle coordinate N0The gray value of place pixel：
I(N0)=0
Obtain image C.
It is described to image C using area growth algorithm obtained in step 4) in step 5), obtain image LiIt is specific
Method can be：
(1) position according to left auricle of heart in image C selects a pixel as seed point s, setting in image C
When the seed point of input meets following threshold values seed point set seed is added in point s by threshold value thresholding：
I0(s)∈[lower,upper]
Wherein, I0(s) indicate that the gray value of seed point s in image C, lower indicate that the lower limit of gray value, upper indicate
The upper limit of gray value；
(2) to each si∈ seed calculates siEight connected region U, by all elements U in eight connected regioni∈ U meets
Seed point set seed is all added in the point of given threshold condition in above-mentioned steps (1), repeats step (2) until not new point is added
Seed point set seed obtains image Li。
In step 6), with the image L obtained in the step 5)iFor initial markup image, phase is carried out to image sequence B '
The Atlas segmentation of neighbour's registration, the specific method for obtaining initial segmentation sequence L can be：
(1) serial number of the image Bi in sequence image B is denoted as i, and next image in sequence image B is denoted as Bi+1, by Bi
The image L obtained with step 5)iLabel is denoted as to (Bi, Li), by BiAs floating image, Bi+1Image is carried out for fixed image to match
Standard obtains transformation parameter fi+1.Image registration using elastix tool (S.Klein, M.Staring, K.Murphy,
M.A.Viergever,J.P.W.Pluim,"elastix:atoolboxforintensitybasedmedical image
registration,"IEEE TransactionsonMedicalImaging,vol.29,no.1,pp.196-205,
January2010 it) realizes, specific step is as follows：
(a) pyramid model is used, by image Bi, Bi+1It is divided into three accuracy classes；
(b) by image Bi, Bi+1It is mapped in logical coordinates from physical coordinates by interpolation method；
(c) with the mode of stochastical sampling respectively from image Bi, Bi+1Middle several sample points of selection, it is whole as representing
Sample point set Si, Si+1；
(d)Si, Si+1Between similarity indicated using mutual information；
(e) optimization algorithm being registrated uses adaptive gradient descent algorithm；
(f) use multi-level registration strategies, successively with rigid transformation, affine transformation, B-spline transformation (D.Rueckert,
L.I.Sonoda,C.Hayes,D.L.G.Hill,M.O.Leach,and
D.J.Hawkes.Nonrigidregistrationusing free-formdeformations:
ApplicationtobreastMRimages.IEEETrans.Med.Imag.,18(8):712-721,1999) three kinds of transformation sides
Formula obtains transformation parameter fi+1。
(2) by transformation parameter fi+1Act on tag image Li, obtain Bi+1Tag image Li+1；
(3) by Bi+1With Li+1As new label to (Bi+1, Li+1), it repeats step (1) and (2), until terminating；
(4) by image BiAs an image B upper in floating image, with sequence image Bi-1According to side described in step (1)
Method carries out image registration, wherein Bi-1For fixed image, transformation parameter f is obtainedi-1；
(5) by transformation parameter fi-1Act on tag image Li, obtain Bi-1Tag image Li-1；
(6) by Bi-1With Li-1As new label to (Bi-1, Li-1), it repeats step (4) and (5), until terminating.
In step 7), the sequence image B in step 2), calculate all foreground points to background dot approximate Euclidean
Distance, the specific method for obtaining range image sequence D can be：(referring to document：Akmal B M,Maragos P.Optimum
design of chamfer distance transforms.[J].IEEE Transactions on Image
Processing A Publication of the IEEE Signal Processing Society,1998,7(10):
1477-84.)：
(1) each image in sequence image B is handled as follows：Range conversion template is pressed from left to right from upper
Sequence sliding under and successively calculates each pixel to the minimum euclidean distance of background, obtains intermediate distance image sequence Bd；
(2) to intermediate distance image sequence BdEach image be handled as follows：Range conversion template is pressed from the right side
To left sequence sliding from bottom to top, each pixel is successively calculated to the minimum euclidean distance of background, obtains distance feature image
Sequence D.
In step 8), the profile of the initial segmentation sequence L that step 6) is obtained is as initial profile, by step 7)
Obtained in range image sequence D as characteristic image, left auricle of heart CT sequence is carried out to sequence image B using quick level set algorithm
The accurate segmentation of column image, specific method can be for (referring to documents：Anon.Insight segmentation and
registration Toolkit[EB/OL].[2008-07-06].http://www.itk.org.)：
(1) each image Di (i=1,2,3 ... ..n) in characteristic image sequence D of adjusting the distance, wherein n is distance feature
The sum of image sequence D is calculated using sigmoid function, forms the input feature vector DS of quick level set algorithmi(i=1,
2,3 ... ..n), wherein sigmoid function is：
Wherein, I indicates image Di, Max, Min are image D respectivelyiMiddle maximum, minimum gradation value, α indicate image DiBrightness
The width (window width) of range, β indicate the brightness (window position) at range center, and I ' indicates output image, i.e. DSi；All DSi(i=
1,2,3 ... ..n) it is quick level set algorithm characteristic sequence, it is denoted as DS；
(2) using the sequence L obtained in step 6) as the initial profile sequence of quick level set algorithm, in step 8) a)
The DS obtained in the process is as quick level set algorithm characteristic sequence, to every image Li, i=1,2,3 ... the ..n in sequence L,
Li is arranged to the zero level collection an of high-dimension function by level set algorithm, this high-dimension function is called level set function：Y (X, t),
Then level set function movement becomes a differential equation, by extracting zero level collection G ((X), t)={ Y (X, t) from output
=0 } come the profile moved, the update of a level set equation computing differential solution of equation ψ is used：
Wherein, A (x) indicates that advection coefficient, P (x) indicate propagation coefficient (being also the coefficient of expansion), and Z (x) indicates curvature
The space regulator coefficient of mean value, α, β, γ indicate the weight of each parameter, are constants.
The present invention carries out left auricle of heart image segmentation to image for the left auricle of heart feature of image of boundary information missing, right first
Image sequence carries out noise reduction process, and reducing noise bring influences；Then in order to obtain initial profile Model sequence, first to individual
Image obtains individual initial segmentation image comprising skeleton pattern information by the boundary information of supplement missing；Again by this individual
The corresponding gray level image of initial segmentation image forms initial Atlas label pair, is obtained by being first registrated the Atlas algorithm divided again
Obtain sequence of partitions；Atlas flag sequence is the initial segmentation sequence for including initial profile model, successively inputs profile information
Level set algorithm obtains its corresponding segmentation result sequence.The present invention is finally reached the purpose being split to image left auricle of heart,
The left auricle of heart that segment boundary lacks is separated from the left ventriculography picture being attached thereto.
For individual left auricle of heart initial segmentation process by priori knowledge, building is adhered the boundary missing of parts of images missing, then
Individual initial segmentation is obtained using partitioning algorithm；Using improved Atlas algorithm from individual initial segmentation image to initial segmentation
Sequence image, it is similar using adjacent image in sequence image, meet requirement of the Atlas algorithm to label centering grayscale image, uses
The adjacent grayscale image in position is registrated individual initial segmentation grayscale image obtained in the sequence with it, by the transformation results of registration
Atlas label (initial segmentation image) is acted on, the Atlas label of neighboring gradation image is obtained, and is combined into new Atlas mark
Note pair, the iteration step can form complete Atlas flag sequence；Accurate Segmentation replaces grayscale image to obtain using distance map
The differentiation information of left auricle of heart and atrium sinistrum, then by the quick level set algorithm of initial profile information input, it is finally reached Accurate Segmentation
Purpose.
Compared with the existing technology, beneficial effects of the present invention are：
1, the present invention is directed to left auricle of heart CT characteristics of image, carries out information supplement to boundary lack part using priori knowledge,
The outgrowth defect for compensating for the algorithm of region growing of connection thresholding successfully obtains first initial point comprising profile information
It cuts.
2, the present invention uses improved Atlas algorithm, now combines the corresponding grayscale image of individual initial segmentation image
For initial markers pair, the grayscale image for recycling the grayscale image of just label centering adjacent with its sequence location is registrated, will be registrated
Transformation results act on Atlas label (initial segmentation image), obtain the Atlas label of neighboring gradation image, and be combined into
New Atlas label pair, the iteration step can form complete Atlas flag sequence, reach automatic obtain and believe comprising profile
The purpose of the initial segmentation sequence of breath.
3, improved Atlas algorithm used in the present invention is using the correlation between sequence, solves similar marker to obtaining
Difficult problem is taken, the segmentation precision of traditional list Atlas algorithm is not only increased, is also greatly reduced in more Atlas algorithms and matches
Quasi-, fusion time complexity.
4, the present invention replaces gray level image to increase location information using range image, and set algorithm of improving the standard is in weak boundary point
Cut the Boundary Recognition ability in problem；And the precision that segmentation is improved by the profile diagram that improved Atlas algorithm obtains is combined,
Realize the Accurate Segmentation of left auricle of heart sequence.
5, the present invention increases the boundary information lacked in left auricle of heart CT sequence image by three steps, improves segmentation essence
Degree, these methods complement each other, all linked with one another, be used in combination with can more accurate Ground Split go out left auricle of heart image, have
The advantages of segmentation precision height, strong robustness.
The present invention is directed to the left auricle of heart feature of image of boundary information missing, and point multiple steps are handled, analyzed, reach from
Precisely divide the purpose of left auricle of heart image in CT image.Individual label of the invention obtains, Atlas obtains initial profile sequence,
Distance feature obtains, the several steps of quick level-set segmentation can functionally complement each other, and single label, which obtains, mainly utilizes priori
The boundary information of information combination priori knowledge supplement missing, recycles the algorithm of region growing of connection thresholding to obtain segmentation, algorithm
Simply, time complexity is low；Improved Atlas algorithm obtains initial profile sequence, for level set algorithm provide as close possible to
The initial profile of real border, reduces the time complexity of level set algorithm, and has prompted the boundary position information of missing, mentions
High segmentation accuracy；Distance feature can improve the problem of boundary information deficiency in gray level image, and set algorithm of improving the standard is boundless
The accuracy of profile at boundary improves robustness；Quick level set algorithm can actively generate closed outline, and in initial profile information
And with the help of distance feature, accurately segmentation result sequence image is generated.Relative to the immediate prior art, have bright
Aobvious distinguishing characteristics, and have the advantages that obvious：Method of the invention can generate sequences segmentation as a result, practical, also not
Need excessive manual intervention, it is only necessary to which the i.e. producible accurately left auricle of heart sequences segmentation of three points of artificial selection is as a result, accurate
It is more superior in property and robustness.
Detailed description of the invention
Fig. 1 is basic flow chart of the invention；
Fig. 2 is one of the left auricle of heart CT image sequence of the embodiment of the present invention；
Fig. 3 is the range conversion template of the embodiment of the present invention；
Fig. 4 is one of the final segmentation result image sequence of the embodiment of the present invention.
Specific embodiment
Present invention is further described in detail with reference to the accompanying drawings and examples, and embodiment is according to flow chart shown in Fig. 1
It explains in detail.
The present invention includes the following steps：
1) left auricle of heart CT sequence image A is inputted；
2) noise reduction process is carried out to the sequence image A in step 1), obtains sequence image B；
3) selecting step 2) in sequence image B in individual CT scheme Bi, residue sequence image is denoted as B '；
4) the image B according to obtained in step 3)iThe position feature information of middle left auricle of heart manually chooses two characteristic points,
The boundary information of left auricle of heart missing is constructed, image C is obtained；
5) to image C using area growth algorithm obtained in step 4), image L is obtainedi；
6) the image L obtained with step 5)iFor initial markup image, the image B ' Atlas for carrying out adjacent registration is divided,
Obtain initial segmentation sequence L.
7) to the sequence image B in step 2), all foreground points is calculated to the approximate Euclidean distance of background dot, obtain distance
Image sequence D；
8) profile for the initial segmentation sequence L for obtaining step 6) is as initial profile, by distance obtained in step 7)
Image sequence D carries out the accurate of left auricle of heart CT sequence image to sequence image B as characteristic image, using quick level set algorithm
Segmentation.
One of left auricle of heart CT image sequence of the embodiment of the present invention is referring to fig. 2.
In step 2), the sequence image A carries out noise reduction process, and the specific method for obtaining sequence image B can be：
(1) image of i-th image after t iterative filtering is denoted as I in sequence image At,i, use curvature anisotropy
Diffusion equation recalculates：
Wherein, c (| x |) indicates specific conductance, wherein▽ expression takes gradient, and t indicates the number of iterations, I
(x, y, i) indicates i-th image in sequence image A；
In the present embodiment, the value of parameters is as follows：C=5, t=4.
(2) the process n times of iterative step (1), i-th image after obtaining noise reduction；
(3) it is incremented by i, enables i=i+1, execute step (1), obtains the image sequence B after retaining edge noise reduction；
In step 3), individual CT schemes B in the selection sequence image BiSpecific method be：
(1) the image number n in left auricle of heart CT sequence image A is calculated；
(2) i-th image is chosen as figure Bi, wherein i is the maximum integer no more than n/2；
In the present embodiment, the value of parameters is as follows：N=60, i=30 choose the 30th image.
In step 4), two characteristic points, building left auricle of heart missing are manually chosen from the image Bi obtained in step 3)
Boundary information, obtain image C specific method be：
(1) two seed points manually chosen are set as p1(x1, y1)、p2(x2, y2), this two o'clock can determine a logical coordinates
Straight line in systemIn physical coordinates system,It can be by from point p1To point p2The pixel fitting extended one by one is constituted, then point p1's
Next pixel position is expressed as：
Wherein, N0Indicate the position of next pixel, N1, N2Indicate position candidate, d1, d2Respectively indicate N1, N2To logic
The distance of next point on coordinate；
(2) according to the positional relationship straight line in left auricle of heart anatomical structureSlope k meet：K ∈ (- 45 °, 0 °) ∪
(- 90 °, -45 °), then its corresponding N1, N2It is expressed as：
(3) when the slope k of straight line meets k ∈ (- 45 °, 0 °)：
Wherein, pi+1Indicate the selection criteria of the next position, dx indicates horizontal departure value in logic, and dy is indicated in logic
Vertical missing value.According to pi+1Value can determine the result of next coordinate：
Work as straight lineSlope k when meeting k ∈ (- 90 °, -45 °)：
Wherein, pi+1Indicate the selection criteria of the next position, dx indicates horizontal departure value in logic, and dy is indicated in logic
Vertical missing value.According to pi+1Value can determine the result of next coordinate：
(4) image B is modifiediMiddle coordinate N0The gray value of place pixel：
I(N0)=0
Obtain image C；
In the present embodiment, the value of parameters is as follows：
P1 (x1, x2)=(115,66), p2 (x2, y2)=(166,110)
It is described to image C using area growth algorithm obtained in step 4) in step 5), obtain image LiIt is specific
Method is：
(1) position according to left auricle of heart in image C selects a pixel as seed point s, setting in image C
When the seed point of input meets following threshold values seed point set seed is added in point s by threshold value thresholding：
I0(s)∈[lower,upper]
Wherein, I0(s) indicate that the coordinate value of seed point s in image C, lower indicate that the lower limit of gray value, upper indicate
The upper limit of gray value.
(2) to each si∈ seed calculates siEight connected region U, by all elements U in eight connected regioni∈ U meets
Seed point set seed is all added in the point of given threshold condition in above-mentioned (1), repeats (2) until seed point set is added in not new point
Seed, the image L of acquisitioni；
In the present embodiment, the value of parameters is as follows：Lower=137, upper=255, s=(152,85).
In step 6), with the image L obtained in the step 5)iFor initial markup image, phase is carried out to image sequence B '
The Atlas segmentation of neighbour's registration, the specific method for obtaining initial segmentation sequence L are：
(1) serial number of the image Bi in sequence image B is denoted as i, and next image in sequence image B is denoted as Bi+1.By Bi
The image L obtained with step 5)iLabel is denoted as to (Bi, Li), by BiAs floating image, Bi+1Image is carried out for fixed image to match
Standard obtains transformation parameter fi+1.Image registration using elastix tool (S.Klein, M.Staring, K.Murphy,
M.A.Viergever,J.P.W.Pluim,"elastix:atoolboxforintensitybasedmedical image
registration,"IEEE TransactionsonMedicalImaging,vol.29,no.1,pp.196-205,
January2010 it) realizes, specific step is as follows：
(a) pyramid model is used, by image Bi, Bi+1It is divided into three accuracy classes；
(b) by image Bi, Bi+1It is mapped in logical coordinates from physical coordinates by interpolation method；
(c) use the mode of stochastical sampling from separated image Bi, Bi+1Middle several sample points of selection, it is whole as representing
Sample point set Si, Si+1；
(d)Si, Si+1Between similarity indicated using mutual information (referring to document：P.Th′
evenazandM.Unser.Optimizationofmutualinformationformultiresolutionimageregist
ration.IEEETrans.Image Process.,9(12):2083–2099,2000.)；
Mutual information can be expressed as：
Wherein, LFAnd LMIt is the set at aturegularaintervals intensity distribution center, p is discrete joint probability, and pFAnd pMPoint
It is not the edge discrete probabilistic of the fixation and moving image by summing to m and f.Joint probability uses B-spline Parzen
Window is estimated：
Wherein, ωFAnd ωMIndicate fixed and mobile B-spline Parzen window.Scaling constants σFAnd σMEqual to by LFAnd LM
The intensity case width of definition.These are directly according to IFAnd IMIntensity value ranges and the quantity of histogram case specified of user | LF
| and | LM|；
(e) optimization algorithm being registrated uses adaptive gradient descent algorithm (bibliography：S.Klein,
J.P.W.Pluim,M.Staring,and M.A.Viergever.Adaptive stochastic gradient descent
optimisationfor image registration.International Journal ofComputer Vision,81
(3):227–239,March 2009)；
(f) multi-level registration strategies are used, successively convert (bibliography with rigid transformation, affine transformation, B-spline：
D.Rueckert,L.I.Sonoda,C.Hayes,D.L.G.Hill,M.O.Leach,and
D.J.Hawkes.Nonrigidregistrationusing free-form deformations:Application to
breast MR images.IEEE Trans.Med.Imag.,18(8):712-721,1999.) three kinds of mapping modes, are become
Change parameter fi+1。
First using the position between rigid transformation adjustment floating image and fixed image：
Wherein, R (x) indicates spin matrix, and C indicates rotation center, and t indicates original transformation parameter group, μ expression parameter to
Amount, by Eulerian angles (rad) and translation set of vectors at being made of length for 3 vector μ, wherein θzIt indicates around perpendicular to figure
The rotation of the axis of picture, tx、tyRespectively indicate the translation on the direction x, y；
It reuses affine transformation and adjusts global shape：
Wherein, C indicates that rotation center, t indicate original transformation parameter group, μ expression parameter vector, and A (x) indicates can be achieved flat
It moves, rotation, the matrix of scaling and shearing, parameter vector μ is by matrix element aijWith translation vector tx、tyComposition, tx、tyTable respectively
Show the translation on the direction x, y；
The deformation of part is finally carried out using B-spline transformation：
Wherein, xkIndicate control point, β3(x) multidimensional B-spline multinomial (M.Unser.Splines three times is indicated:
Aperfect fit for signal and image processing.IEEE Signal Process.Mag.,16(6):
22-38,1999.), pkIt indicates B-spline coefficient vector (control point displacement), σ indicates B-spline control point spacing, NxIt indicates by by B
The point set of x point control constraints in batten；Control point mesh is defined σ=(σ by the amount of space between control point1..., σd) (its
Middle d indicates picture size), each direction can be different；The quantity P=(P1 ..., Pd) at control point passes through M=
(P1 × ... × Pd) × d determines the quantity of parameter M；Parameter vector is as follows：μ=(p1x, p2x..., pP1, p1y, p2y...,
pP2)T；To obtain transformation parameter fi+1。
(2) by transformation parameter fi+1Act on tag image Li, obtain Bi+1Tag image Li+1；
(3) by Bi+1With Li+1As new label to (Bi+1, Li+1), it repeats step (1) and (2), until terminating；
(4) by image BiAs an image B upper in floating image, with sequence image Bi-1According to side described in step (1)
Method carries out image registration, wherein Bi-1For fixed image, transformation parameter f is obtainedi-1；
(5) by transformation parameter fi-1Act on tag image Li, obtain Bi-1Tag image Li-1；
(6) by Bi-1With Li-1As new label to (Bi-1, Li-1), it repeats step (4) and (5), until terminating；
In the present embodiment, the value of parameters is as follows：N=2048, N=3, Iter (Iterr, Itera, IterB)=
(800,1200,1000), | LF|=| LM|=(16,32,64)
In step 7), the sequence image B in step 2), calculate all foreground points to background dot approximate Euclidean
Distance, the specific method for obtaining range image sequence D are：(referring to document：Akmal B M,Maragos P.Optimum
design of chamfer distance transforms.[J].IEEE Transactions on Image
Processing A Publication of the IEEE Signal Processing Society,1998,7(10):
1477-84.)：
(1) each image in sequence image B is handled as follows：Range conversion template is pressed from left to right from upper
Sequence sliding under and successively calculates each pixel to the minimum euclidean distance of background, obtains intermediate distance image sequence Bd；
(2) to intermediate distance image sequence BdEach image be handled as follows：Range conversion template is pressed from the right side
To left sequence sliding from bottom to top, each pixel is successively calculated to the minimum euclidean distance of background, obtains distance feature image
Sequence D；
In the present embodiment, range conversion template is as shown in figure 3, wherein the value of parameters is as follows：A=5, b=7, c=
11；
In step 8), the profile of the initial segmentation sequence L that step 6) is obtained is as initial profile, by step 7)
Obtained in range image sequence D as characteristic image, left auricle of heart CT sequence is carried out to sequence image B using quick level set algorithm
The accurate segmentation of column image, specific method are (referring to document：Anon.Insight segmentation and
registration Toolkit[EB/OL].[2008-07-06].http://www.itk.org.)：
(1) each image Di (i=1,2,3 ... ..n) in characteristic image sequence D of adjusting the distance, wherein n is distance feature
The sum of image sequence D is calculated using sigmoid function, forms the input feature vector DS of quick level set algorithmi(i=1,
2,3 ... ..n), wherein sigmoid function is：
Wherein I indicates image Di, Max, Min are image D respectivelyiMiddle maximum, minimum gradation value, α indicate image DiBrightness model
The width (window width) enclosed, β indicate the brightness (window position) at range center, and I ' indicates output image, i.e. DSi；All DSi(i=1,
2,3 ... ..n) it is quick level set algorithm characteristic sequence, it is denoted as DS；
(2) using the sequence L obtained in step 6) as the initial profile sequence of quick level set algorithm, in step 8) a)
The DS obtained in the process as quick level set algorithm characteristic sequence, in sequence L every image Li (i=1,2,3 ...
..n), Li is arranged to the zero level collection an of high-dimension function by level set algorithm, this high-dimension function is called level set function：Y
(X, t), then level set function movement becomes a differential equation, by extracting zero level collection G ((X), t)={ Y from output
(X, t)=0 } come the profile that is moved.Use the update of a level set equation computing differential solution of equation ψ：
Wherein, A (x) indicates that advection coefficient, P (x) indicate propagation coefficient (being also the coefficient of expansion), and Z (x) indicates curvature
The space regulator coefficient of mean value, α, β, γ indicate the weight of each parameter, are constants；
In the present embodiment, the value of parameters is as follows：The quick level set that P (x) is obtained using (1) process in step 8)
Algorithm characteristics sequence D S.α, β, γ are set as the default value in ITK development kit (referring to document：Anon.Insight
segmentation and registration Toolkit[EB/OL].[2008-07-06].http://
www.itk.org.).The final result images for obtaining left auricle of heart CT image segmentation, as shown in Figure 4.
Claims (8)
1. a kind of left auricle of heart CT image partition method, it is characterised in that include the following steps：
1) left auricle of heart CT sequence image A is inputted；
2) noise reduction process is carried out to the sequence image A in step 1), obtains sequence image B；
3) selecting step 2) in sequence image B in individual CT scheme Bi, residue sequence image is denoted as B '；
4) individual CT according to obtained in step 3) schemes BiThe position feature information of middle left auricle of heart manually chooses two characteristic points, structure
The boundary information of left auricle of heart missing is built, image C is obtained；
5) to image C using area growth algorithm obtained in step 4), image L is obtainedi；
6) the image L obtained with step 5)iFor initial markup image, the image B ' Atlas for carrying out adjacent registration is divided, is obtained
Initial segmentation sequence L；
7) to the sequence image B in step 2), all foreground points is calculated to the approximate Euclidean distance of background dot, obtain range image
Sequence D；
8) profile for the initial segmentation sequence L for obtaining step 6) is as initial profile, by range image obtained in step 7)
Sequence D carries out precisely dividing for left auricle of heart CT sequence image to sequence image B as characteristic image, using quick level set algorithm
It cuts.
2. a kind of left auricle of heart CT image partition method as described in claim 1, it is characterised in that in step 2), the sequence chart
As A progress noise reduction process, the specific method for obtaining sequence image B is：
(1) image of i-th image after t iterative filtering is denoted as I in sequence image At,i, use curvature anisotropy parameter
Equation recalculates：
Wherein, c (| x |) indicates specific conductance,▽ expression takes gradient, and t indicates the number of iterations, I (x, y, i) table
Show i-th image in sequence image A；
(2) the process n times of iterative step (1), i-th image after obtaining noise reduction；
(3) it is incremented by i, enables i=i+1, execute step (1), obtains the image sequence B after retaining edge noise reduction.
3. a kind of left auricle of heart CT image partition method as described in claim 1, it is characterised in that in step 3), the selection step
It is rapid 2) in sequence image B in individual CT scheme BiSpecific method be：
(1) the image number n in left auricle of heart CT sequence image A is calculated；
(2) i-th image is chosen as figure Bi, wherein i is the maximum integer no more than n/2.
4. a kind of left auricle of heart CT image partition method as described in claim 1, it is characterised in that described according to step in step 4)
It is rapid 3) obtained in individual CT scheme BiThe position feature information of middle left auricle of heart, manually chooses two characteristic points, and building left auricle of heart lacks
The boundary information of mistake, the specific method for obtaining image C are：
(1) two seed points manually chosen are set as p1(x1, y1)、p2(x2, y2), this two o'clock determines straight in a logical coordinates system of pulse train
Line l, in physical coordinates system, l is by from point p1To point p2The pixel fitting extended one by one is constituted, then point p1Next picture
Vegetarian refreshments position is expressed as：
Wherein, N0Indicate the position of next pixel, N1, N2Indicate position candidate, d1, d2Respectively indicate N1, N2Onto logical coordinates
The distance of next point；
(2) met according to the slope k of the positional relationship straight line l in left auricle of heart anatomical structure：K ∈ (- 45 °, 0 °) ∪ (-
90 °, -45 °), then its corresponding N1, N2It is expressed as：
(3) when the slope k of straight line meets k ∈ (- 45 °, 0 °)：
Wherein, pi+1Indicate the selection criteria of the next position, dx indicates horizontal departure value in logic, and dy indicates vertical in logic
Deviation, according to pi+1Value determine the result of next coordinate：
When the slope k of straight line l meets k ∈ (- 90 °, -45 °)：
Wherein, pi+1Indicate the selection criteria of the next position, dx indicates horizontal departure value in logic, and dy indicates hanging down in logic
Straight deviation.According to pi+1Value determine the result of next coordinate：
(4) image B is modifiediMiddle coordinate N0The gray value of place pixel：
I(N0)=0
Obtain image C.
5. a kind of left auricle of heart CT image partition method as described in claim 1, it is characterised in that described to step in step 5)
4) image C using area growth algorithm obtained in obtains image LiSpecific method be：
(1) position according to left auricle of heart in image C selects a pixel as seed point s, given threshold in image C
When the seed point of input meets following threshold values seed point set seed is added in point s by thresholding：
I0(s)∈[lower,upper]
Wherein, I0(s) indicate that the gray value of seed point s in image C, lower indicate that the lower limit of gray value, upper indicate gray value
The upper limit；
(2) to each si∈ seed calculates siEight connected region U, by all elements U in eight connected regioni∈ U, meets step
(1) seed point set seed is all added in the point of given threshold condition in, repeats step (2) until seed point set is added in not new point
Seed obtains image Li。
6. a kind of left auricle of heart CT image partition method as described in claim 1, it is characterised in that in step 6), with from step 5)
Obtained in image LiFor initial markup image, the image sequence B ' Atlas for carrying out adjacent registration is divided, initial segmentation is obtained
The specific method of sequence L is：
(1) serial number of the image Bi in sequence image B is denoted as i, and next image in sequence image B is denoted as Bi+1, by BiWith step
The rapid image L 5) obtainediLabel is denoted as to (Bi, Li), by BiAs floating image, Bi+1Image registration is carried out for fixed image,
Obtain transformation parameter fi+1；Image registration realizes that specific step is as follows using elastix tool：
(a) pyramid model is used, by image Bi, Bi+1It is divided into three accuracy classes；
(b) by image Bi, Bi+1It is mapped in logical coordinates from physical coordinates by interpolation method；
(c) with the mode of stochastical sampling respectively from image Bi, Bi+1Middle several sample points of selection, as the sample point for representing entirety
Collect Si, Si+1；
(d)Si, Si+1Between similarity indicated using mutual information；
(e) optimization algorithm being registrated uses adaptive gradient descent algorithm；
(f) multi-level registration strategies are used, successively three kinds of mapping modes is converted with rigid transformation, affine transformation, B-spline, obtains
Obtain transformation parameter fi+1；
(2) by transformation parameter fi+1Act on tag image Li, obtain Bi+1Tag image Li+1；
(3) by Bi+1With Li+1As new label to (Bi+1, Li+1), it repeats step (1) and (2), until terminating；
(4) by image BiAs an image B upper in floating image, with sequence image Bi-1According to method described in step (1) into
Row image registration, wherein Bi-1For fixed image, transformation parameter f is obtainedi-1；
(5) by transformation parameter fi-1Act on tag image Li, obtain Bi-1Tag image Li-1；
(6) by Bi-1With Li-1As new label to (Bi-1, Li-1), it repeats step (4) and (5), until terminating.
7. a kind of left auricle of heart CT image partition method as described in claim 1, it is characterised in that described to step in step 7)
2) the sequence image B in calculates all foreground points to the approximate Euclidean distance of background dot, obtains the specific of range image sequence D
Method is：
(1) each image in sequence image B is handled as follows：Range conversion template is pressed from left to right from top to bottom
Sequence sliding, successively calculate each pixel arrive background minimum euclidean distance, acquisition intermediate distance image sequence Bd；
(2) to intermediate distance image sequence BdEach image be handled as follows：Range conversion template is pressed from right to left certainly
Sequence sliding on down successively calculates each pixel to the minimum euclidean distance of background, obtains distance feature image sequence D.
8. a kind of left auricle of heart CT image partition method as described in claim 1, it is characterised in that described by step in step 8)
6) profile of the initial segmentation sequence L obtained is as initial profile, using range image sequence D obtained in step 7) as feature
Image, the accurate segmentation of left auricle of heart CT sequence image is carried out using quick level set algorithm to sequence image B, and specific method is：
(1) each image Di, i=1,2 in characteristic image sequence D of adjusting the distance, 3 ... ..n, wherein n is distance feature image
The sum of sequence D is calculated using sigmoid function, forms the input feature vector DS of quick level set algorithmi, i=1,2,
3 ... ..n, wherein sigmoid function be：
Wherein, I indicates image Di, Max, Min are image D respectivelyiMiddle maximum, minimum gradation value, α indicate image DiBrightness range
Width, β indicate range center brightness, I ' indicate output image, i.e. DSi；All DSiFor quick level set algorithm feature
Sequence is denoted as DS, wherein i=1,2,3 ... ..n；
(2) a) process using the sequence L obtained in step 6) as the initial profile sequence of quick level set algorithm, in step 8)
The DS of middle acquisition is as quick level set algorithm characteristic sequence, and to every image Li, i=1,2 in sequence L, 3 ... ..n are horizontal
Li is arranged to the zero level collection an of high-dimension function by set algorithm, this high-dimension function is called level set function：Y (X, t), then
Level set function movement becomes a differential equation, by extracting zero level collection G ((X), t)={ Y (X, t)=0 } from output
The profile moved uses the update of a level set equation computing differential solution of equation ψ：
Wherein, A (x) indicates that advection coefficient, P (x) indicate that propagation coefficient, Z (x) indicate the space regulator system of curvature mean value
Number, it is constant that α, β, γ, which indicate the weight of each parameter,.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810553854.0A CN108876769B (en) | 2018-05-31 | 2018-05-31 | Left auricle CT image segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810553854.0A CN108876769B (en) | 2018-05-31 | 2018-05-31 | Left auricle CT image segmentation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108876769A true CN108876769A (en) | 2018-11-23 |
| CN108876769B CN108876769B (en) | 2020-11-03 |
Family
ID=64336197
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810553854.0A Expired - Fee Related CN108876769B (en) | 2018-05-31 | 2018-05-31 | Left auricle CT image segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108876769B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109767396A (en) * | 2019-01-04 | 2019-05-17 | 北京朗视仪器有限公司 | A kind of oral cavity CBCT image de-noising method based on image dynamic partition |
| CN110175984A (en) * | 2019-04-17 | 2019-08-27 | 杭州晟视科技有限公司 | A kind of separation method of model, device, terminal and computer storage medium |
| CN112070752A (en) * | 2020-09-10 | 2020-12-11 | 杭州晟视科技有限公司 | Method, device and storage medium for segmenting auricle of medical image |
| CN112244883A (en) * | 2020-09-10 | 2021-01-22 | 北京思创贯宇科技开发有限公司 | Method and system for extracting left auricle data parameters based on CT image |
| CN112308845A (en) * | 2020-11-03 | 2021-02-02 | 赛诺威盛科技（北京）有限公司 | Left ventricle segmentation method and device and electronic equipment |
| CN112862827A (en) * | 2021-04-26 | 2021-05-28 | 杭州晟视科技有限公司 | Method, device, terminal and storage medium for determining opening parameters of left atrial appendage |
| CN116363160A (en) * | 2023-05-30 | 2023-06-30 | 杭州脉流科技有限公司 | CT perfusion image brain tissue segmentation method and computer equipment based on level set |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101852768A (en) * | 2010-05-05 | 2010-10-06 | 电子科技大学 | Workpiece flaw identification method based on compound characteristics in magnaflux powder inspection environment |
| CN101916443A (en) * | 2010-08-19 | 2010-12-15 | 中国科学院深圳先进技术研究院 | Processing method and system of CT image |
| US20130083983A1 (en) * | 2011-10-04 | 2013-04-04 | Siemens Aktiengesellschaft | Method and System for Segmentation and Removal of Pulmonary Arteries, Veins, Left Atrial Appendage |
| CN103400365A (en) * | 2013-06-26 | 2013-11-20 | 成都金盘电子科大多媒体技术有限公司 | Automatic segmentation method for lung-area CT (Computed Tomography) sequence |
-
2018
- 2018-05-31 CN CN201810553854.0A patent/CN108876769B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101852768A (en) * | 2010-05-05 | 2010-10-06 | 电子科技大学 | Workpiece flaw identification method based on compound characteristics in magnaflux powder inspection environment |
| CN101916443A (en) * | 2010-08-19 | 2010-12-15 | 中国科学院深圳先进技术研究院 | Processing method and system of CT image |
| US20130083983A1 (en) * | 2011-10-04 | 2013-04-04 | Siemens Aktiengesellschaft | Method and System for Segmentation and Removal of Pulmonary Arteries, Veins, Left Atrial Appendage |
| CN103400365A (en) * | 2013-06-26 | 2013-11-20 | 成都金盘电子科大多媒体技术有限公司 | Automatic segmentation method for lung-area CT (Computed Tomography) sequence |
Non-Patent Citations (1)
| Title |
|---|
| 黄韫栀等: ""左心耳图像分割方法研究"", 《四川大学学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109767396A (en) * | 2019-01-04 | 2019-05-17 | 北京朗视仪器有限公司 | A kind of oral cavity CBCT image de-noising method based on image dynamic partition |
| CN110175984A (en) * | 2019-04-17 | 2019-08-27 | 杭州晟视科技有限公司 | A kind of separation method of model, device, terminal and computer storage medium |
| CN110175984B (en) * | 2019-04-17 | 2021-09-14 | 杭州晟视科技有限公司 | Model separation method and device, terminal and computer storage medium |
| CN112070752A (en) * | 2020-09-10 | 2020-12-11 | 杭州晟视科技有限公司 | Method, device and storage medium for segmenting auricle of medical image |
| CN112244883A (en) * | 2020-09-10 | 2021-01-22 | 北京思创贯宇科技开发有限公司 | Method and system for extracting left auricle data parameters based on CT image |
| CN112308845A (en) * | 2020-11-03 | 2021-02-02 | 赛诺威盛科技（北京）有限公司 | Left ventricle segmentation method and device and electronic equipment |
| CN112862827A (en) * | 2021-04-26 | 2021-05-28 | 杭州晟视科技有限公司 | Method, device, terminal and storage medium for determining opening parameters of left atrial appendage |
| CN112862827B (en) * | 2021-04-26 | 2021-07-16 | 杭州晟视科技有限公司 | Method, device, terminal and storage medium for determining opening parameters of left atrial appendage |
| CN116363160A (en) * | 2023-05-30 | 2023-06-30 | 杭州脉流科技有限公司 | CT perfusion image brain tissue segmentation method and computer equipment based on level set |
| CN116363160B (en) * | 2023-05-30 | 2023-08-29 | 杭州脉流科技有限公司 | CT perfusion image brain tissue segmentation method and computer equipment based on level set |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108876769B (en) | 2020-11-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108876769A (en) | A kind of left auricle of heart CT image partition method | |
| Shimizu et al. | Segmentation of multiple organs in non-contrast 3D abdominal CT images | |
| WO2021088747A1 (en) | Deep-learning-based method for predicting morphological change of liver tumor after ablation | |
| CN110503649B (en) | Liver segmentation method based on spatial multi-scale U-net and superpixel correction | |
| US8150119B2 (en) | Method and system for left ventricle endocardium surface segmentation using constrained optimal mesh smoothing | |
| Zhu et al. | Automatic segmentation of the left atrium from MR images via variational region growing with a moments-based shape prior | |
| CN111524170B (en) | Pulmonary CT image registration method based on unsupervised deep learning | |
| CN108010048B (en) | Multi-atlas-based hippocampus segmentation method for automatic brain MRI (magnetic resonance imaging) image | |
| CN109214397A (en) | The dividing method of Lung neoplasm in a kind of lung CT image | |
| CN105976364B (en) | Based on the statistical average model building method for simplifying weighted-graph | |
| CN109509193B (en) | Liver CT atlas segmentation method and system based on high-precision registration | |
| CN107563323A (en) | A kind of video human face characteristic point positioning method | |
| CN107220965A (en) | A kind of image partition method and system | |
| Sung et al. | A unified gradient-based approach for combining ASM into AAM | |
| US11334995B2 (en) | Hierarchical systems and methods for image segmentation | |
| WO2022247218A1 (en) | Image registration method based on automatic delineation | |
| CN111080658A (en) | Cervical MRI image segmentation method based on deformable registration and DCNN | |
| CN114882051A (en) | Automatic segmentation and three-dimensional reconstruction method for pelvic bone tumor based on multi-modal image | |
| CN111724389B (en) | Method, device, storage medium and computer equipment for segmenting CT image of hip joint | |
| CN113362341A (en) | Air-ground infrared target tracking data set labeling method based on super-pixel structure constraint | |
| CN110570430B (en) | Orbital bone tissue segmentation method based on volume registration | |
| CN110610491A (en) | Liver tumor region segmentation method of abdominal CT image | |
| CN109829885B (en) | Method for automatically identifying primary tumor of nasopharyngeal carcinoma based on deep semantic segmentation network | |
| CN107240114B (en) | A kind of semi-automatic medical image cutting method based on distance function shape constraining | |
| CN107093176A (en) | A kind of head mri image partition method based on atlas |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201103 |