CN103793910B - A kind of dividing method of heterogencity medical image - Google Patents

A kind of dividing method of heterogencity medical image Download PDF

Info

Publication number
CN103793910B
CN103793910B CN201410028478.5A CN201410028478A CN103793910B CN 103793910 B CN103793910 B CN 103793910B CN 201410028478 A CN201410028478 A CN 201410028478A CN 103793910 B CN103793910 B CN 103793910B
Authority
CN
China
Prior art keywords
seed point
background
image
point group
foreground seeds
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.)
Expired - Fee Related
Application number
CN201410028478.5A
Other languages
Chinese (zh)
Other versions
CN103793910A (en
Inventor
陈海斌
周凌宏
甄鑫
王琳婧
肖阳
胡洁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southern Medical University
Original Assignee
Southern Medical University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southern Medical University filed Critical Southern Medical University
Priority to CN201410028478.5A priority Critical patent/CN103793910B/en
Publication of CN103793910A publication Critical patent/CN103793910A/en
Application granted granted Critical
Publication of CN103793910B publication Critical patent/CN103793910B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Image Processing (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

The present invention relates to a kind of dividing method of heterogencity medical image, the method comprises the steps of：Foreground seeds point and background seed point are first selected on image to be split；Then, the half-tone information according to the seed point set selecting, estimates the probability that each gray scale belongs to prospect or background in image to be split, and is mapped to each picture element of image, obtains corresponding probability density distribution figure；Then more respectively with the foreground seeds point of selection and background seed point for growing seed point, with a probability threshold value as growth conditionss on corresponding probability density distribution figure, execution algorithm of region growing, obtains the foreground seeds point group after automatic growth and background seed point group；Finally, use the seed point group after the automatic growth obtaining as the seed point of Random Walk Algorithm, execute Random Walk Algorithm, obtain last segmentation result.The method of the invention can reduce the quantity to initial seed point and location sen-sitivity, significantly improves the segmentation precision of heterogencity medical image.

Description

A kind of dividing method of heterogencity medical image
Technical field
The present invention relates to graphical analyses are and in particular to the dividing method of medical image.
Background technology
Developing rapidly with medical imaging, medical image segmentation for clinical diagnosis and treatment all have important Meaning.Current image segmentation algorithm is broadly divided into manual segmentation, Interactive Segmentation and full-automatic dividing three class.Manual segmentation ratio Relatively time-consuming, and require divider to have rich experience.Full-automatic dividing, without manual intervention, is generally relatively specific for simply equal The segmentation of even image, but the segmentation for complicated and diversified medical image, the precision of generally segmentation does not reach the need of clinic Ask.And Interactive Segmentation only need to add a small amount of manual intervention, just can automatically be partitioned into preferable result.In order to meet complexity The segmentation of image requires, usual Interactive Segmentation using relatively broad.
It is one of most popular interactive image segmentation algorithm at present based on the image segmentation algorithm of graph theory, wherein wrap Include Random Walk Algorithm (Random walker), figure segmentation algorithm (Graph cut) and shortest path first (Shortest Paths) etc..The feature of such algorithm is exactly to need to select a few class seed points on image by user, then according to seed point The image information providing, all pixels point in image is referred to all kinds of seed points that user selectes, realize image final point Cut.Such algorithm is effectively improved the precision of image segmentation, but such method treats the gray scale of target internal in segmentation figure picture Difference is very sensitive, and it may erroneous segmentation be different classifications that the gray scale difference of target can lead to the pixel of different gray scales in target.Non- equal One property image is then precisely the presence of such gray feature, thus leading to the method poor to the Target Segmentation effect of heterogencity.Though The seed that so this problem can be selected by rolling up user counts out to improve, but such method is little in seed point In the case of do not simply fail to obtain accurate segmentation result, and the change of segmentation result is very tight with the change of initial seed point Weight, so this necessarily leads to the efficiency split to decline.The segmentation of especially heterogencity objective is it would be desirable to user is in target Each aspect all select seed point, this is very unrealistic.It is true that heterogencity image is common figure clinically Picture, such as, many organization of human body Various Tissues are constituted, and tumor zones of different assumes different half-tone informations etc. in the picture, this The target resulting in image to be split a bit assumes the feature of heterogencity.Generally such algorithm is in the segmentation of heterogencity image The result of satisfaction cannot be obtained.
A kind of image segmentation algorithm [Mu Ke based on Mean Shift and random walk that Mu Ke et al. announced in 2012 With (2012) such as Cheng Wei. " image segmentation algorithm based on Mean Shift and random walk. " Liaoning Technical University's journal:Natural Science version 32 (1):27-30.], the method comprises the steps of：(1) foreground seeds point and background seed point are selected；(2) adopt With Mean Shift algorithm, pretreatment is carried out to image, divide an image into some homogeneous regions, with homogeneous region as node Carry out random walk；(3) and then, using the weights between mahalanobis distance definition region, seed point is improved, increased auxiliary Help seed point；(4) carry out random walk using the seed point of auxiliary seed point and user's mark, carry out the merging of homogeneous region, Realize the final segmentation of image.Above method inhibits the interference of noise to a certain extent, weakens user's mark seed point The impact to random walk segmentation result for the position and number, be the segmentation result that user obtains satisfaction.But the method is only It is as node using homogeneous region, assists the generation of seed point to also rely on homogeneous region, so being non-equal in target to be split During one property, the non-homogeneous region of target area is not still enabled to split well, still cannot realize accurately splitting.
Content of the invention
The technical problem to be solved is to provide a kind of dividing method of heterogencity medical image, and the method can Significantly improve the segmentation precision of heterogencity medical image.
The technical solution that the present invention solves the above problems is：
A kind of dividing method of heterogencity medical image, the method comprises the following steps：
(1) user chooses foreground seeds point in the target area on image to be split, chooses background kind outside target area Sub-, and obtain the half-tone information of foreground seeds point and background seed point；
(2) according to the half-tone information obtained by step (1), foreground seeds point and background seed point are allowed respectively by the following method Carry out automatic growth：
(A) adopt lower formula (I) and the method for the Multilayer networks shown in (II), estimate the picture element of each gray value Distribution probability P (Ii/gs),
In formula (I) and (II), IiRepresent the gray value of picture element；S represents the classification of seed point, that is, prospect class seed point or Background classes seed point, gsRepresent the classification of seed point set, i.e. prospect class seed point set or background classes seed point set；tqFor seed point Collection gsIn q-th seed point；σ is Multilayer networks variance, σ=0.5 × DI, wherein DIFor prospect class seed point set and the back of the body The meansigma methodss of the gray scale difference between scape class seed point set；MaxI is the maximum gradation value of image to be split,P is gray value, ZsFor the normalized parameter in Multilayer networks computing；
(B) the distribution probability P that the Bayes's condition probability formula according to lower formula (III) and step (A) are tried to achieve (Ii/gs), Schilling,And lower formula (III) is transformed to lower formula (IV), then by P (Ii/gs) substitute into lower formula (IV), ask Solve the distribution probability that each gray value in image to be split belongs to a class in prospect class or background classes, and by image to be split The gray value mapping of each picture element obtains image to be split and corresponds to such probability density distribution figure, then, according to prospect The complementary relationship of class and background classes probability density distribution figure tries to achieve another kind of probability density distribution figure,
G in upper formula (III) and (IV)1Expression prospect class seed point set, g2Represent background classes seed point set；
(C) with seed point set gsAs the seed point of region growing method, take threshold parameter in 0.5～1To be more than This threshold parameterAs growth conditionss, in probability density distribution figure PsOn carry out region growing, obtain foreground seeds point group and Background seed point group；
(3) the foreground seeds point group obtained by step (2) and background seed point group are processed：Respectively by foreground seeds Point group and background seed point group carry out down-sampled, or, only background seed point group is reconstructed and foreground seeds point group is constant； Wherein, the reconstructing method of described background seed point group is：First calculate the geometric center of foreground seeds point group, then before calculating and comprising this The border circular areas of scape seed point group, the point on this border circular areas edge is the background seed point group after reconstructing；
(4) and then, the foreground seeds point group that will process through step (3) and background seed point group are as the prospect of random walk Seed point and background seed point, are split to image to be cut using random walk method, obtain segmentation result.
Foreground seeds point group described in above-mentioned steps (3) and background seed point group carry out down-sampled and background seed point group It is reconstructed in order that foreground seeds point group and background seed point group rarefaction, to accelerate splitting speed.
Random Walk Algorithm described in above-mentioned steps (4) is a kind of classical Interactive Segmentation algorithm, by Leo Grady (Grady, L., Random Walks for Image Segmentation.Pattern Analysis and is proposed first Machine Intelligence,IEEE Transactions on,2006.28(11):P.1768-1783.), the method will Piece image treats as a width non-directed graph, and this figure is made up of summit and side, each of each vertex correspondence original image pixel Point, the weights in every a line represent between two summits the probability size mutually reaching, and described weights are by the two of two vertex correspondence Intensity difference between individual picture element determines.Then user passes through labelling initial seed point, solves remaining picture element in image and corresponds to Summit reaches the probability of labelling seed for the first time, and each picture element is ranged the first seed point classification reaching maximum probability, Thus segmentation is realized to image.Above-mentioned algorithm has stronger weak boundary segmentation ability and anti-noise ability.
The invention has the advantages that：
(1) by Multilayer networks method, the heterogencity of target gray is converted into the homogeneity of probability, and will have The pixel automatic growth having likelihood probability is seed point, thus achieving the Accurate Segmentation to heterogencity target well.
(2) present invention dexterously utilizes the foreground seeds point of user's mark and the half-tone information of background seed point, estimates Each of image gray value belongs to the probability of prospect, and widely different gray scale institute in each classification in making image of all categories The probability accounting for becomes close to, thus the gray scale heterogencity of image is converted into the homogeneity of probability, and being close with gained Probability as restrictive condition, foreground seeds point and the position of background seed point and number are expanded, thus reducing Dependency for seed point.
(3) method of the present invention all enables accurate point to the target of homogeneity and the non-target to homogeneity Cut.
Brief description：
Fig. 1：The abdominal CT images of lumbar vertebra to be split.
Fig. 2：The lung CT image of the lumbar vertebra to be split of labelling seed point set.
Fig. 3：The abdominal CT images corresponding prospect probability density distribution figure of lumbar vertebra to be split.
Fig. 4：The superimposed image of the abdominal CT images of the seed point group after automatic growth and lumbar vertebra to be split.
Fig. 5：The superimposed image of the abdominal CT images of the seed point group after down-sampled and lumbar vertebra to be split.
Fig. 6：Segmentation result figure using lumbar vertebra in the abdominal CT images that the method for the invention obtains.
Fig. 7：Segmentation result figure using lumbar vertebra in the abdominal CT images that existing method obtains.
Fig. 8：Pelvic cavity CT image containing apparatus for wave-energy source.
Fig. 9：The pelvic cavity CT image containing apparatus for wave-energy source of labelling seed point set
Figure 10：Foreground seeds point group after automatic growth and the background seed point group after reconstruct and the pelvic cavity containing apparatus for wave-energy source The superimposed image of CT image.
Figure 11：Segmentation result figure using apparatus for wave-energy source in the pelvic cavity CT image that the method for the invention obtains.
Figure 12：Segmentation result figure using apparatus for wave-energy source in the pelvic cavity CT image that existing method obtains.
Specific embodiment：
Example 1 (segmentations of heterogencity two-dimensional medical images)
Describe as a example abdominal CT (Computed Tomography) image of certain patient shown in by Fig. 1 for the present embodiment The implementation process of the method for the invention.The size of Fig. 1 is 512 × 512, wherein needs the lumbar vertebra split to belong to heterogencity Target, this target includes highdensity cortical bone, the compact bone of intermediate density and low-density bone marrow.Concrete dividing method is such as Lower described：
Step 1：Read in CT image as shown in Figure 1, by MATLAB GUI graphical interfaces, select to represent lumbar vertebra Foreground seeds point and the background seed point representing other normal structures of abdominal part.The seed point of selection is superimposed upon on Fig. 1, obtain as The abdominal CT images of the lumbar vertebra to be split of labelling seed point set shown in Fig. 2, obtaining in figure black arrow indication lines is mark Note foreground seeds point and in figure white arrow indication lines be labelling background seed point half-tone information.
Step 2：According to the half-tone information of step 1 gained, allow foreground seeds point and background seed point respectively by the following method Carry out automatic growth：
(2.1) half-tone information according to the seed point set selecting in step 1, close using the probability shown in formula (I) and (II) The method that degree is estimated, estimates distribution probability P (I in foreground and background for each gray valuei/g1) and P (Ii/g2)；
(2.2) by the P obtaining (Ii/g1) and P (Ii/g2) substitute into formula (IV) respectively, obtain before each gray value belongs to Distribution probability P (the g of scape1/Ii).Then, first obtained as Fig. 3 institute by the gray value mapping of each picture element of image to be split The abdominal CT images corresponding prospect probability density distribution figure of the lumbar vertebra to be split showing, further according to prospect probability density distribution figure Complementary relationship and background probability density profile between, obtains corresponding background probability density profile；
(2.3) the foreground seeds point set g being selected with step 11As growth seed point, with the threshold parameter more than 0.6 As growth conditionss, region growing is carried out on Fig. 3, using the picture element in the growth district obtaining as after automatic growth before Scape seed point group；Again with the background seed point set g of step 1 labelling2As growth seed point, with the threshold parameter more than 0.6 As growth conditionss, the abdominal CT images corresponding background probability density profile of lumbar vertebra to be split carries out region life Long, using the picture element in the growth district obtaining as the background seed point group after automatic growth.By the foreground seeds after expanding Point group and background seed point group are superimposed upon the seed point group after obtaining automatic growth as shown in Figure 4 on Fig. 1 and lumbar vertebra to be split The superimposed image of the abdominal CT images of bone.
Step 3:Foreground seeds point group obtained by step (2) and background seed point group are carried out down-sampled process, its tool Body step is as follows：
The foreground seeds point group and background seed point group obtaining after automatic growth is carried out on the coordinate of image to be split Down-sampled, obtain down-sampled after foreground seeds point group and background seed point group, with the seed point group after down-sampled with to be split The abdominal CT images of lumbar vertebra are overlapped obtaining Fig. 5；Above-mentioned down-sampled method is that the sampling interval of foreground seeds point group is 1 Pixel unit, the sampling interval of background seed point group is 5 pixel units.
Step 4：Down-sampled rear foreground seeds point group and background seed point group will be obtained as random walk in step 3 Foreground seeds point and background seed point carry out random walk computing, obtain segmentation result as shown in Figure 6.Target in Fig. 6 The black lines of surrounding are cut-off rule.
It is that the inventive method is compared with prior art, the image shown in Fig. 1 is adopted the method choosing described in step 1 Take foreground seeds point and background seed point, then as Mu Ke et al. in 2012 disclosed in a kind of based on Mean Shift and random The image segmentation calculation based on Mean Shift and random walk for the image partition method [Mu Ke and Cheng Wei etc. (2012). " of migration Method. " Liaoning Technical University's journal:Natural science edition 32 (1):27-30.] split, obtain segmentation result as shown in Figure 7, Black lines around target therein are cut-off rule.Fig. 6 and Fig. 7 is compared the segmentation it is clear that the method for the invention Result is accurate compared with prior art.
Example 2 (segmentations of heterogencity 3 d medical images)
The present embodiment with shown in Fig. 8 certain cervical cancer patient accept closely Afterloading radiotherapy when captured containing The method of the invention is described for heterogencity Three Dimensional Medical Visualization process as a example the pelvic cavity CT image of apparatus for wave-energy source.Fig. 8 Size be 256 × 256 × 55, the apparatus for wave-energy source of in figure belongs to heterogencity objective, and it includes highdensity metal tube matter, The plastics of intermediate density and low-density liquid and air etc..Concrete dividing method is as described below：
Step 1：Read in pelvic cavity CT image as shown in Figure 8, by MATLAB GUI graphical interfaces, selection can represent Shi Yuan The foreground seeds point of device and the background seed point representing other organizational structuries of pelvic cavity.The seed point of selection is superimposed upon on Fig. 8, obtains To the labelling seed point set as shown in Figure 9 pelvic cavity CT image containing apparatus for wave-energy source, acquisition black arrow indication lines are labelling Foreground seeds point and white arrow indication lines be labelling background seed point half-tone information；
Step 2：According to the half-tone information of step 1 gained, allow foreground seeds point and background seed point respectively by the following method Carry out automatic growth：
(2.1) half-tone information according to the seed point set selecting in step 1, using the probability shown in formula (I) and formula (II) The method of density estimation, estimates distribution probability P (I in foreground and background for each gray valuei/g1) and P (Ii/g2)；
(2.2) by the P obtaining (Ii/g1) and P (Ii/g2) substitute into formula (IV) respectively, obtain before each gray value belongs to Probability P (the g of scape1/Ii).Then, the basin containing apparatus for wave-energy source is obtained by the gray value mapping of each picture element of image to be split Chamber CT image corresponding prospect probability density distribution figure.Further according to prospect probability density distribution figure and background probability density profile Between complementary relationship, obtain corresponding background probability density profile；
(2.3) the foreground seeds point set g being selected with step 11As growth seed point, with the threshold parameter more than 0.9 As growth conditionss, the pelvic cavity CT image corresponding prospect probability density distribution figure containing apparatus for wave-energy source carries out region growing, Using the picture element in the growth district obtaining as the foreground seeds point group after automatic growth；Again with the initial kind of step 1 labelling Sub- point set g2As growth seed point, with the threshold parameter more than 0.9As growth conditionss, in the pelvic cavity CT containing apparatus for wave-energy source Carry out region growing, using the picture element in the growth district obtaining as automatic on image corresponding background probability density profile Background seed point group after growth.
Step 3：Background seed point group after the automatic growth that step 2 is obtained is reconstructed, and it comprises the following steps that：
(3.1) obtain prospect kind in the geometric center on each transverse section and this transverse section for the foreground seeds point group Son point reaches the ultimate range of geometric center.
(3.2) with the geometric center on each transverse section as the center of circle, to reach 1.2 times of distances of the ultimate range at center For radius, obtain a circle, using the point on this circle as the background seed point group on this transverse section.Using Amide software integration Obtain the background seed point group after foreground seeds point group and reconstruct after automatic growth as shown in Figure 10 and the basin containing apparatus for wave-energy source The superimposed image of chamber C T image, wherein white arrow indication region are the background seed point group after reconstruct, black arrow indication area Domain is the foreground seeds point group after automatic growth.
Step 4：The background seed point as random walk for the background seed point group being obtained with step 3, before being obtained with step 2 Scape seed point group be random walk foreground seeds point, segmentation figure picture is treated using random walk method and is split, obtain as Segmentation result shown in Figure 11.In Figure 11, the gray area of white arrow indication is to be superimposed upon the pelvic cavity CT image containing apparatus for wave-energy source On segmentation mask image.
It is that the inventive method is compared with prior art, the image shown in Fig. 7 is adopted the method choosing described in step 1 Take foreground seeds point and background seed point, then press one kind that Mu Ke et al. announced in 2012 and be based on Mean Shift and random The image segmentation calculation based on Mean Shift and random walk for the image partition method [Mu Ke and Cheng Wei etc. (2012). " of migration Method. " Liaoning Technical University's journal:Natural science edition 32 (1):27-30.] three-dimensional expanding method split, obtain as Figure 12 Shown segmentation result.In Figure 12, the gray area of white arrow indication is to be superimposed upon on the pelvic cavity CT image containing apparatus for wave-energy source The mask image of segmentation.It can be seen that the method for the invention is in the segmentation of three-dimensional heterogencity medical image, compared with prior art essence Really.

Claims (1)

1. a kind of dividing method of heterogencity medical image, the method comprises the following steps：
(1) user chooses foreground seeds point in the target area on image to be split, chooses background seed point outside target area, And obtain the half-tone information of foreground seeds point and background seed point；
(2) according to the half-tone information obtained by step (1), foreground seeds point and background seed point is allowed to carry out by the following method respectively Automatic growth：
(A) adopt lower formula (I) and the method for the Multilayer networks shown in (II), estimate the distribution of the pixel of each gray value Probability P (Ii/gs),
P ( I i / g s ) = 1 Z s Σ t q ∈ g s e - ( I i - t q ) 2 σ - - - ( I )
Z s = Σ p = 0 M a x I Σ t q ∈ g s e - ( p - t q ) 2 σ - - - ( I I )
In formula (I) and (II), IiRepresent the gray value of pixel；S represents the classification of seed point, i.e. prospect class seed point or background Class seed point, gsRepresent the classification of seed point set, i.e. prospect class seed point set or background classes seed point set；tqFor seed point set gs In q-th seed point；σ is Multilayer networks variance, σ=0.5 × DI, wherein DIFor prospect class seed point set and background classes The meansigma methodss of the gray scale difference between seed point set；MaxI is the maximum gradation value of image to be split,P is gray value, ZsFor the normalized parameter in Multilayer networks computing；
(B) the distribution probability P (I that the Bayes's condition probability formula according to lower formula (III) and step (A) are tried to achievei/gs), Shilling,And lower formula (III) is transformed to lower formula (IV), then by P (Ii/gs) substitute into lower formula (IV), solve to be split In image, each gray value belongs to the distribution probability of a class in prospect class or background classes, and each picture by image to be split The gray value mapping of vegetarian refreshments obtains image to be split and corresponds to such probability density distribution figure, then, according to prospect class and background The complementary relationship of class probability density distribution figure tries to achieve another kind of probability density distribution figure,
P ( g s / I i ) = P ( I i / g s ) · P ( g s ) P ( I i / g 1 ) · P ( g 1 ) + P ( I i / g 2 ) · P ( g 2 ) - - - ( I I I )
P ( g s / I i ) = P ( I i / g s ) P ( I i / g 1 ) + P ( I i / g 2 ) - - - ( I V )
G in upper formula (III) and (IV)1Expression prospect class seed point set, g2Represent background classes seed point set；
(C) with seed point set gsAs the seed point of region growing method, take threshold parameter in 0.5～1With more than this threshold Value parameterAs growth conditionss, in probability density distribution figure PsOn carry out region growing, obtain foreground seeds point group and background Seed point group；
(3) the foreground seeds point group obtained by step (2) and background seed point group are processed：Respectively by foreground seeds point group With background seed point group carry out down-sampled, or, only background seed point group is reconstructed and foreground seeds point group is constant；Its In, the reconstructing method of described background seed point group is：First calculate the geometric center of foreground seeds point group, then calculate and comprise this prospect The border circular areas of seed point group, the point on this border circular areas edge is the background seed point group after reconstructing；
(4) and then, the foreground seeds point group that will process through step (3) and background seed point group are as the foreground seeds of random walk Point and background seed point, are split to image to be cut using random walk method, obtain segmentation result.
CN201410028478.5A 2014-01-21 2014-01-21 A kind of dividing method of heterogencity medical image Expired - Fee Related CN103793910B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410028478.5A CN103793910B (en) 2014-01-21 2014-01-21 A kind of dividing method of heterogencity medical image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410028478.5A CN103793910B (en) 2014-01-21 2014-01-21 A kind of dividing method of heterogencity medical image

Publications (2)

Publication Number Publication Date
CN103793910A CN103793910A (en) 2014-05-14
CN103793910B true CN103793910B (en) 2017-03-08

Family

ID=50669533

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410028478.5A Expired - Fee Related CN103793910B (en) 2014-01-21 2014-01-21 A kind of dividing method of heterogencity medical image

Country Status (1)

Country Link
CN (1) CN103793910B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105427272A (en) * 2014-09-17 2016-03-23 富士通株式会社 Image processing device, image processing method and electronic device
CN104637074A (en) * 2015-02-26 2015-05-20 西安理工大学 Method for extracting interest area image of multi-target free-hand sketch
CN104881872B (en) * 2015-05-27 2018-06-26 浙江大学 A kind of optics microangiography image segmentation and evaluation method
CN105701795A (en) * 2015-12-30 2016-06-22 中国科学院深圳先进技术研究院 Method and system for abdominal bone segmentation in medical image
CN105608711A (en) * 2016-01-18 2016-05-25 华东理工大学 Local-sensitive-histogram-based dynamic target tracking and extracting method of video
CN106127785B (en) * 2016-06-30 2019-02-22 重庆大学 Image significance detection method based on manifold ranking and random walk
CN107464250B (en) * 2017-07-03 2020-12-04 深圳市第二人民医院 Automatic breast tumor segmentation method based on three-dimensional MRI (magnetic resonance imaging) image
CN109903266B (en) * 2019-01-21 2022-10-28 深圳市华成工业控制股份有限公司 Sample window-based dual-core density estimation real-time background modeling method and device
CN111815647A (en) * 2020-06-23 2020-10-23 上海硕恩网络科技股份有限公司 Stochastic walk image segmentation algorithm based on toboggan

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101908214B (en) * 2010-08-10 2012-05-23 长安大学 Moving object detection method with background reconstruction based on neighborhood correlation
CN102881021B (en) * 2012-10-25 2016-02-10 上海交通大学 Based on the aorta petal ultrasonic image division method of probability distribution and continuous max-flow

Also Published As

Publication number Publication date
CN103793910A (en) 2014-05-14

Similar Documents

Publication Publication Date Title
CN103793910B (en) A kind of dividing method of heterogencity medical image
Choi et al. Clinical evaluation of atlas-and deep learning-based automatic segmentation of multiple organs and clinical target volumes for breast cancer
US20210290096A1 (en) Methods and systems for segmenting organs in images using a cnn-based correction network
CN107230206A (en) A kind of 3D Lung neoplasm dividing methods of the super voxel sequence lung images based on multi-modal data
CN106780518B (en) A kind of MR image three-dimensional interactive segmentation method of the movable contour model cut based on random walk and figure
CN102920477B (en) Device and method for determining target region boundary of medical image
EP2252204B1 (en) Ct surrogate by auto-segmentation of magnetic resonance images
US8023734B2 (en) 3D general lesion segmentation in CT
Wu et al. Prediction of molecular subtypes of breast cancer using BI-RADS features based on a “white box” machine learning approach in a multi-modal imaging setting
CN104715484B (en) Automatic tumor imaging region segmentation method based on improved level set
CN102496156B (en) Medical image segmentation method based on quantum-behaved particle swarm cooperative optimization
CN106204587A (en) Multiple organ dividing method based on degree of depth convolutional neural networks and region-competitive model
Palma et al. Detection of masses and architectural distortions in digital breast tomosynthesis images using fuzzy and a contrario approaches
Tunçay et al. Realistic microwave breast models through T1-weighted 3-D MRI data
CN104182755A (en) Mammary gland molybdenum target X-ray image block feature extraction method based on tower-shaped principal component analysis (PCA)
Hussain et al. Femur segmentation in DXA imaging using a machine learning decision tree
CN107545579A (en) A kind of cardiac segmentation method, equipment and storage medium
CN107507189A (en) Mouse CT image kidney dividing methods based on random forest and statistical model
CN105389821A (en) Medical image segmentation method based on combination of cloud module and image segmentation
KR20190090663A (en) Method of automatically generating a curve corresponding to an dental arch shape on dental ct slice automatically selected from three dimensional dental ct image
Luan et al. An efficient end-to-end CNN for segmentation of bone surfaces from ultrasound
Shim et al. Fully automated breast segmentation on spiral breast computed tomography images
CN106997592A (en) Fuzzy clustering medical image cutting method with real-time
Smeets et al. Segmentation of liver metastases using a level set method with spiral-scanning technique and supervised fuzzy pixel classification
Ganasala et al. Semiautomatic and automatic brain tumor segmentation methods: performance comparison

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
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: 20170308

Termination date: 20220121