CN106447682A - Automatic segmentation method for breast MRI focus based on Inter-frame correlation - Google Patents
Automatic segmentation method for breast MRI focus based on Inter-frame correlation Download PDFInfo
- Publication number
- CN106447682A CN106447682A CN201610761183.8A CN201610761183A CN106447682A CN 106447682 A CN106447682 A CN 106447682A CN 201610761183 A CN201610761183 A CN 201610761183A CN 106447682 A CN106447682 A CN 106447682A
- Authority
- CN
- China
- Prior art keywords
- image
- segmentation
- level set
- frame
- tumor
- 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.)
- Pending
Links
- 230000011218 segmentation Effects 0.000 title claims abstract description 69
- 238000000034 method Methods 0.000 title claims abstract description 50
- 210000000481 breast Anatomy 0.000 title claims abstract description 20
- 206010028980 Neoplasm Diseases 0.000 claims abstract description 24
- 238000004422 calculation algorithm Methods 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000003709 image segmentation Methods 0.000 claims description 5
- 230000006872 improvement Effects 0.000 claims description 5
- 210000005075 mammary gland Anatomy 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 abstract 1
- 238000007670 refining Methods 0.000 abstract 1
- 230000003902 lesion Effects 0.000 description 7
- 206010006187 Breast cancer Diseases 0.000 description 5
- 208000026310 Breast neoplasm Diseases 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000000877 morphologic effect Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 210000001519 tissue Anatomy 0.000 description 5
- 201000010099 disease Diseases 0.000 description 4
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 4
- 238000004195 computer-aided diagnosis Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000000638 solvent extraction Methods 0.000 description 3
- 238000005481 NMR spectroscopy Methods 0.000 description 2
- 201000008275 breast carcinoma Diseases 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000009977 dual effect Effects 0.000 description 2
- 210000004907 gland Anatomy 0.000 description 2
- 238000009607 mammography Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241000196324 Embryophyta Species 0.000 description 1
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 description 1
- 206010020751 Hypersensitivity Diseases 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 239000008186 active pharmaceutical agent Substances 0.000 description 1
- 208000009956 adenocarcinoma Diseases 0.000 description 1
- 208000026935 allergic disease Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 230000005669 field effect Effects 0.000 description 1
- 239000000945 filler Substances 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000009610 hypersensitivity Effects 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000004630 mental health Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 210000000062 pectoralis major Anatomy 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 210000004872 soft tissue Anatomy 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 210000000779 thoracic wall Anatomy 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/77—Retouching; Inpainting; Scratch removal
-
- 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
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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
- G06T2207/20036—Morphological image processing
-
- 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
- G06T2207/30068—Mammography; Breast
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to an automatic segmentation method for a breast MRI focus based on inter-frame correlation. The method comprises: reading an MRI image; preprocessing the image, and performing coarse segmentation on the preprocessed image to determine an initial contour of a focus; and by adopting an improved C- V level set model method, performing fine segmentation on a coarsely segmented image that is obtained in the prior step, further refining the tumor contour on the basis of a coarsely segmented contour, and optimizing an obtained finely segmented result by combining inter-frame correlation. The method provided by the present invention has higher accuracy and can better segment the focus.
Description
Technical field
The present invention relates to medical image segmentation technology, particularly to a kind of mammary gland magnetic resonance image partition method
Background technology
Breast carcinoma is one of modal malignant tumor of women, seriously threatens women physical and mental health and life.Breast carcinoma
Early discovery diagnosis and treatment be favorably improved the year survival rate and life quality of patient.Mr imaging technique (MRI) has
Generally acknowledged hypersensitivity (more than 90%), strong to the resolution energy of all kinds of soft tissue structures, provide multiple sequences, in multiple directions
Image, be that the clear and definite cause of disease provides organizational structure, the form of focus, the more abundant image information such as lump distribution.Mammary gland
MRI observes to early diagnosing mammary cancer, treatment phase, drafts rational operation plan and the later stage follows the tracks of and has great auxiliary to make
With.
However, due to there is electronic noise, skew field distortion and volume effect during NMR (Nuclear Magnetic Resonance)-imaging, leading to
There is noise and inclined field effect, in breast MRI image, the border between various tissues is often smudgy in image.Additionally, it is logical
Often the diagnosis of breast MRI relies on personal experience and the Professional knowledge of clinician mostly.Computer-aided diagnosises technology, is a kind of
Radiologist can be helped to carry out the emerging technology of examination to medical image.It has when processing a large amount of patient's image data
Very high practicality, is analyzed to certain disease by substantial amounts of medical image processing method and models, lesion information is led to
Cross parameter model to make a distinction with normal structure, screen out real focus, provide the ginseng that mark diagnoses further as doctor
Examine.The appearance of computer-aided diagnosises technology and development, significantly reduce the workload of doctor, and more importantly improve breast
The accuracy rate of adenocarcinoma diagnosis and sensitivity, thus saving the life of ten million patient with breast cancer, simultaneously also to accurate using computer
Identification and segmentation breast lesion are put forward higher requirement.
As one of computer-aided diagnosis in mammography system study hotspot, its main segmentation object has lesion segmentation：
1) segmentation of the pathological changes body of quantitative analyses and follow-up classification task are done for tumor property；2) pertinent tissue structures (skin side
Edge, thoracic wall, heart etc.) automatic profile obtain.For the segmentation of this two class, lot of domestic and international scholar has done correlational study and has obtained
Certain achievement, including based on morphologic dividing method[1-3], based on cluster dividing method[4-6], the segmentation based on edge
Method[7,8]With the dividing method based on particular theory[9,10].But various dividing methods all have its limitation, for example threshold method and
Region growth method is less efficient and more sensitive to noise, FCM and MRF scheduling algorithm excessively relies on data it is impossible to ensure segmentation result
Anatomically correct.Up-to-date has researched and proposed the dividing method that two kinds of algorithms combine, such as with reference to SLIC super-pixel and water
The partitioning algorithm of flat collection[11](SLIC+ and DRLSE), the method is used for the lesion segmentation of mammography X, achieves preferably
Segmentation effect. the area grayscale yet with MRI image is uneven, and experimental result on MRI image for this algorithm is unsatisfactory.
First, this algorithm adopt simple linear Iterative Clustering (simple linear iterative clustering,
SLIC the super-pixel shape difference) producing is larger, can produce impact to subsequent singulation.Thin segmentation step employs apart from regularization
Level Set Method (DRLSE), but the method will artificially determine the symbol of constant Evolution Rates according to the position of initial curve, needs
Want manual intervention it is impossible to realize adaptivenon-uniform sampling.In addition, the method only considered two dimension segmentation, do not make full use of breast MRI
The 3-dimensional image spatial information that scanning provides.On MRI image, especially when initial less with tumor on abort frame and gray scale and week
When enclosing tissue and being closer to, the experimental result of the method is unsatisfactory, or even cannot be partitioned into focus.Some combine frame-to-frame correlation
Document[12-13]Although combining inter-frame information, however it is necessary that doctor marks manually, be not suitable for processing mass data.
Content of the invention
For the above-mentioned deficiency of prior art, the present invention provides one kind can be independent of manually marking, automatic accurate segmentation
The initial dividing method with the breast MRI focus of breast lesion on abort frame, the present invention, according to the feature of breast MRI image, changes
Enter existing C-V level set algorithm, the SLIC0 super-pixel based on frame-to-frame correlation and C-V level set combined, coarse segmentation and
Subdivision is cut and is combined, and adopts SLIC clustering algorithm in coarse segmentation part；Combine improved C-V level set mould in thin partitioning portion
Type, more rapidly and accurately determines tumor's profiles, improves segmentation precision；The related letter of three-dimensional interframe is combined in two-dimentional cutting procedure
Breath, improves stability and the sensitivity of segmentation, and the 3-D solid structure for showing focus lays the first stone.Technical scheme is as follows：
A kind of automatic division method of the breast MRI focus based on frame-to-frame correlation includes：
A. read MRI image；
B. pretreatment image, including following two steps：
C. to image coarse segmentation after pretreatment, determine the initial profile of focus, comprise the following steps：
D. it is finely divided using coarse segmentation image I (x, y) that improved C-V Level Set Models method obtains to step C and cut,
Refine tumor's profiles further on the basis of coarse segmentation profile, improvements are as follows：
1) C-V Level Set Models are by minimizing energy function, determining final segmentation contour, but its energy function mould
Only utilize global information in type, and cannot correctly split the uneven image of Luminance Distribution, in the energy function of C-V Level Set Models
Middle addition local energy item, improves the energy of image segmentation, is k with a window size with the statistical information of the gray scale of local
The average convolution operator of × k, does convolution mathematic interpolation to image I (x, y), so that the intensity distribution difference of image reduces, thin
Tumor's profiles are accurately found in segmentation step；
2) it is the convergence rate improving C-V Level Set Models, cut efficiency to improve subdivision, in the energy of C-V Level Set Models
Introduce energy penalty term in flow function, do gradient calculation to according to the level set function φ of closed curve C construction, and to Grad
Square integrate again, to accelerate calculating speed；
3) it is the smoothness controlling zero level set function, evolution length of a curve compensation term is added to C-V level set mould
In the energy function of type, integral operation is done to the gradient product of Dirac function and level set function, it is to avoid in thin segmentation result
Isolated zonule occurs；
E. the thin segmentation result obtaining step D, in conjunction with frame-to-frame correlation, is optimized, comprises the following steps：
1) the suspected tumor regions area of each frame obtaining after elaborate division by calculation cuts, automatically chooses tumor area largest frames and makees
For iteration key frame, obtain the tumor's profiles R of key frame；
2) using R as initial profile, using the above-mentioned improved C-V Level Set Models upper and lower frame to key frame respectively
Image carries out segmentation adjustment, obtains more accurate segmentation result, then with this result for initial profile to forward and backward iteration, Zhi Daojian
Do not detect more tiny profile, you can think and in this frame, do not comprise lump.
Preferably,
Pretreatment image in B, including following two steps：
1) intercept ROI, take out the minimum image rectangular area comprising mammary gland；
2) morphology enhancing is carried out using top cap (top-hat) computing, obtain image after pretreatment.
To image coarse segmentation after pretreatment, determine the initial profile of focus, comprise the following steps：
1) super-pixel segmentation：Super-pixel segmentation is carried out using SLIC clustering algorithm；
2) suspicious region screening, using by the super-pixel region merging technique of close gray value, only retains the higher area of gray value
Domain, as the method for suspicious region, is screened, and obtains the image after coarse segmentation.
Using the present invention, pretreated breast MRI image is carried out with super-pixel coarse segmentation and C-V segments after cutting, then
Split in conjunction with frame-to-frame correlation and optimized, compared to single dividing method, accuracy rate is higher, even if MRI image initiates
With the focus very little on abort frame, focus gray value and background gray levels situation relatively, also can preferably be partitioned into disease
Stove, the 3 D stereo for follow-up focus shows, has laid good basis.Meanwhile, whole cutting procedure is full-automatic dividing, reduces
Man-machine interactively.
Brief description
Fig. 1：Partitioning algorithm system block diagram based on frame-to-frame correlation.
Fig. 2：The breast MRI image being obtained by magnetic resonance scanner.
Fig. 3：The result that each step produces：3 (a) is the automatic ROI image intercepting；3 (b) is the knot after Morphological scale-space
Really；3 (c) is SLIC0 super-pixel segmentation result；3 (d) is the thin segmentation result of level set.
Fig. 4：The segmentation result to one complete sequence of breast MRI lesion image for the present invention, the numerical value below every little figure
It is its actual picture sequence numbers in High-resolution MRI sequence.
Specific embodiment
With reference to shown in Fig. 1, wherein comprise step performed below：Read MRI image 10 first；Then gained image is carried out
Pretreatment 20；Next to image coarse segmentation 30 after pretreatment, determine the initial profile of focus；To the image subdivision after coarse segmentation
Cut 40, refine focus edge；Finally adopt the segmentation 50 based on frame-to-frame correlation, improve segmentation accuracy rate.
Read breast MRI image 10 in above-mentioned steps, the image of acquisition is as shown in Figure 2.The harvester of above-mentioned image
For sharp Pu Intera Achieva 1.5T magnetic resonance scanner.Fat suppression sequence is dynamically strengthened using the quick volume acquisition of axle position
Row (dyn_eTHRIVE) scan, and the relevant parameter of imaging is：Repetition time TR=4.4ms, echo time TE=2.2ms, upset
FA=12 ° of angle.Thickness 2mm, FOV is 100 × 100cm, and gradation of image is 12, and matrix size is 352 × 352 (units：Picture
Element), multiple scanning 8 times, each dynamic scan time is 60s.
Step 20 carries out pretreatment to the MRI image obtaining, and comprises the following steps that：
Automatically intercept ROI21：Due to smaller for the size relative to view picture image for the size of breast lesion, thus want from
Move and select the ROI comprising focus, make focus be located at the center of ROI as far as possible.The operation choosing ROI makes cutting object have more pin
To property, heart etc. can be excluded there is the impact to cutting procedure for the tissue being remarkably reinforced, reduce the complexity of cutting procedure.
The position of the breast being obtained due to same equipment is basically unchanged it is possible to set a rectangular window, including whole
Breast image, removes the parts such as heart.During automatic selection ROI, the rectangular window size of ROI can be actual according to focus
Size adjustment.For convenience of statistical analysiss segmentation result, in addition to huge focus individually, remaining focus present invention is used uniformly across size
For the ROI window of 80 × 80 pixels, this size is much larger than the size of general focus.
Morphology strengthens 22：
Compared with normal image, MRI initial pictures are easily subject to the shadow of image documentation equipment and formation condition or other compositions
Ring, the quality of image it is possible that the situation degenerated, it could even be possible to artefact occurs.In order to solve these problems it is necessary to right
Image carries out pretreatment, removes the noise of image, strengthens the contrast of image.Invention applies a kind of dual morphology side
Method.Effectively background area can be suppressed, and prominent Probability Area, lay a solid foundation for segmentation detection afterwards.
Morphological scale-space method includes opening and closing operation and top cap computing.Morphology opening operation is used for deletion and does not comprise structural elements
The region of element, so that the contour smoothing of destination object, disconnects narrow link, removes tiny projection.Morphologic closed operation
It is primarily used to tiny cavity in filler body, connect the object closing on.Select suitable structural element, image is first opened
After computing, closed operation can effectively remove picture noise.The selection of structural element directly determines that morphology removes the effect of noise.
In Morphological scale-space method, another very conventional method is top cap (top-hat) computing, and it can strengthen
Image detail.Original image is deducted the result images through opening operation by top cap computing.Top cap computing can strengthen the moon of image
Shadow details, detects the peak signal of image.Top cap operational formula is as follows：
The comprising the following steps that of this Enhancement Method：
F (i, j) represents artwork, B1Represent first selected structural element, then first artwork is carried out with a Top-
Hat computing：
Now r1(i, j) is exactly artwork and artwork and structural element B1The difference image of image after opening operation.Here to select
Structural element B1Size slightly larger than lump, then the body of gland group that size in image is more than lump just can be removed by this step
Knit, retain size suitable with lump and be less than lump region.
If another B2Represent second selected structural element, in order to remove small-sized noise region and other little chi
Very little useless region is in addition it is also necessary to carry out an opening operation again to the result obtaining：
This dual morphology operations, eliminate undersized noise impurity and large-sized gland tissue, only respectively
Retain the region in the possible magnitude range of lump in image.This method not only can project doubtful lump region, can also exempt from
Go the step removing pectoralis major in pretreatment for avoiding interference in some are based on threshold method.
Coarse segmentation step 30, mainly includes following two operations：
Super-pixel segmentation 31, using feature between pixel similarity degree by group pixels, the redundancy of image can be obtained
Information, reduces the complexity of successive image process task to a great extent.The present invention adopts simple linear iteraction cluster
I.e. zero parameter version of SLIC clustering algorithm carries out coarse segmentation.
1) initialization seed point：Assume that image has N number of pixel, pre-segmentation is the super-pixel of K same size, then every
The size of individual super-pixel is N/K, and the distance of neighboring seeds point is approximatelyIn order to avoid seed point is in image
Marginal position and follow-up cluster process is interfered, need to move seed point in 3 × 3 windows centered on itself
To the position that Grad is minimum, distribute a single label for each seed point simultaneously.
2) measuring similarity：For each pixel, calculate the similarity degree between seed point adjacent thereto respectively, will
The label of most like seed point is assigned to this pixel；Constantly this process of iteration, until convergence.Define (lk,ak,bk,xk,yk) be
Five dimension coordinates of arbitrfary point, (l in Lab spacei,ai,bi,xi,yi) for seed point coordinate figure, then the measurement relation of similarity
For：
DSIt is the sum of the Lab distance and space pixel distance normalized cumulant on mesh spacing.dlabRepresent Lab color
Distance, dxyRepresenting space length, S is maximum space distance in class. maximum Lab color distance was both different with picture difference,
Different with cluster difference, so taking fixed constant m that a span is [Isosorbide-5-Nitrae 0] to replace.M is bigger, and space length is similar
Impact in cluster process for the property is bigger.In order to improve the arithmetic speed of algorithm, SLIC algorithm is clustering to each seed point
When, only in the 2S × 2S region centered on seed point, search for similar pixel point, rather than find in whole image.
SLIC0 algorithm is the zero parameter version of SLIC.In SLIC algorithm, m and S is single constant, and SLIC0 algorithm
Then the maximum space and color distance that each iteration of dynamically normalization obtains, the space as next iteration and color away from
From.Distance after improvement and formula are：
mlabAnd mxyFor the maximum color that obtains in front an iteration and space length.As shown from the above formula, SLIC0 is not
Need arrange parameter m and S again, algorithm, by the Texture complication according to image zones of different, dynamically adjusts suitable parameters value, makes
SLIC0 algorithm in the case of keeping speed constant, the more regular unification of produced super-pixel, advantageously reduce follow-up place
The impact to algorithm for super-pixel shape difference during reason.
Suspicious region screens after 32, SLIC0 draws initial segmentation result, the super-pixel of gained will be screened, stays
Suspicious region.Due to the gray value of background area super-pixel visible significantly lower than lump region, and morphology strengthen back scenic spot
The gray value of domain super-pixel is significantly lower than the gray value of original image same area, and lump region super-pixel shape is more regular,
Class circularity is higher.Therefore the super-pixel region merging technique of close gray value only can be retained the higher region of gray value as can
Doubtful region.
40 are cut to the image subdivision after coarse segmentation, tumor's profiles are refined using C-V Level Set Method.Chan and Vese proposes
C-V model be optimization to Mumford-Shan energy functional, promote drilling of level set curve using inside and outside gray average
Change, divisible non-flanged is fuzzy, no gradient meaning image, and to insensitive for noise.This model hypothesis domain of definition is the figure of Ω
It is divided into target inside (C) and two homogeneous regions of background outside (C) as I (x, y) is closed curve C.Its energy functional
It is expressed as follows:
Wherein μ, ν are each term coefficient, all negated negative value；c1,c2It is respectively the average in target and background region.φ be set to according to
According to the level set function of closed curve C construction, Ω is the length of closed curve C, and φ is the gradient of φ, obtains level set function
Expression formula is：
Wherein, define regularization Heaviside function H (φ (x, y)) and Dirac function δ (φ (x, y)) is：
To profile C derivation, the partial differential equation obtaining representing with level set function φ are
φ (0, x, y)=φ0(x,y)
Wherein：T express time,ForGradient produce unit vector field divergence, its meaning is non-plane motion
When unit volume rate of change.c1And c2Calculating formula is
C-V model mainly solves the image segmentation problem not having limbus, but it has two kinds of major defects：1) only sharp
With global information it is impossible to correctly split the uneven image of Luminance Distribution.2) inefficient in the calculation, required time is relatively
Long.For overcoming both major defects, C-V model is improved.
Global information is only used based on C-V model, and this has been asked cannot correctly to split the uneven image of Luminance Distribution
Topic, to the addition local entity in C-V model.Local energy item is used the statistical information of the gray scale of local to improve image segmentation
Energy, local energy function E (d1,d2, C) it is described as follows:
Wherein, gkFor the average convolution operator for k × k for the window size.* it is that volume calculates son, d1,d2It is error image respectively
(gk* I (x, y)-I (x, y)) gray average in target and background region.
The calculating time that second major defect of C-V model is inefficient in calculating process, required is considerably long,
Relatively time consuming, solution is introduced into energy penalty term P (φ)：
For seeking gradient signs, φ is to seek gradient to φ.
In addition, in order to control the smoothness of zero level set function and avoid that isolated zonule occurs in segmentation result,
Evolution length of a curve compensation term is added in adjustment energy term.Length compensation item is defined as follows：
Regularization Heaviside function H (φ (x, y)) and Dirac function δ (φ (x, y)) are defined as above.
C-V model after improvement, it based on the energy functional of level set is：
Wherein
In the present invention, parameter is set to:λ1=1, λ2=1, μ=2, υ=1.5, α=0.04, ε=1.0.C-V after improvement
Algorithm has the curve evolvement model advantage no reinitializing, and faster, and convergence is more preferably for convergence rate, and segmentation result is more
Press close to tumor's profiles.
Using the segmentation of frame-to-frame correlation, first have to choose intermediate frame.The intermediate frame that conventional method adopts needs to cure mostly
Shi Shoudong marks, and cutting procedure is not full-automatic.For reducing the needs of man-machine interaction, improve detection efficiency, set forth herein one
Plant the algorithm automatically choosing intermediate frame.Result after being cut according to subdivision calculates the area of suspected abnormality in each frame, will comprise disease
The maximum frame of the area of stove regards as intermediate frame, to constrain sagittal segmentation.This is because the larger frame of tumor area, swell
Tumor profile is relatively clearly it is easier to obtain accurate segmentation result as the initial profile of iteration.
First, result subdivision cut is optimized, and removes the cut zone of mistake.The maximum gauge of general breast tumor
It is about 30mm, on MRI image, constitute about 250 pixels.And the region area of erroneous segmentation part, generally less than maximum swollen
The 1/5 of tumor area, therefore, removes the connected region that area on every two field picture is less than maximum tumor area 1/5 (i.e. 50 pixel), that is,
The part of erroneous segmentation while retaining maximum tumor, can be removed.
The suspected tumor regions area of each frame that elaborate division by calculation obtains after cutting, chooses tumor area largest frames conduct automatically
Iteration key frame, obtains the tumor's profiles R of key frame.
Using R as initial profile, respectively a upper and lower two field picture of key frame is split using improved C-V model
Adjustment, obtains more accurate segmentation result, then with this result for initial profile to forward and backward iteration.
Until can't detect more tiny profile, you can think and do not comprise lump in this frame.
Claims (3)
1. a kind of automatic division method of the breast MRI focus based on frame-to-frame correlation includes：
A. read MRI image；
B. pretreatment image, including following two steps：
C. to image coarse segmentation after pretreatment, determine the initial profile of focus, comprise the following steps：
D. it is finely divided using coarse segmentation image I (x, y) that improved C-V Level Set Models method obtains to step C and cuts, thick
Tumor's profiles are refined further, improvements are as follows on the basis of segmentation contour：
1) C-V Level Set Models are by minimizing energy function, determining final segmentation contour, but in its energy function model
Only utilize global information, and cannot correctly split the uneven image of Luminance Distribution, add in the energy function of C-V Level Set Models
Enter local energy term, improve the energy of image segmentation with the statistical information of the gray scale of local, be k × k with a window size
Average convolution operator, convolution mathematic interpolation is done to image I (x, y) so that image intensity distribution difference reduce, cut in subdivision
Tumor's profiles are accurately found in step；
2) it is the convergence rate improving C-V Level Set Models, cut efficiency to improve subdivision, in the energy letter of C-V Level Set Models
Introduce energy penalty term in number, do gradient calculation to according to the level set function φ of closed curve C construction, and Grad is put down
Side integrates, to accelerate calculating speed again；
3) it is the smoothness controlling zero level set function, evolution length of a curve compensation term is added to C-V Level Set Models
In energy function, integral operation is done to the gradient product of Dirac function and level set function, it is to avoid occur in thin segmentation result
Isolated zonule；
E. the thin segmentation result obtaining step D, in conjunction with frame-to-frame correlation, is optimized, comprises the following steps：
1) the suspected tumor regions area of each frame obtaining after elaborate division by calculation cuts, chooses tumor area largest frames automatically as repeatedly
For key frame, obtain the tumor's profiles R of key frame；
2) using R as initial profile, using the above-mentioned improved C-V Level Set Models upper and lower two field picture to key frame respectively
Carry out segmentation adjustment, obtain more accurate segmentation result, then with this result for initial profile to forward and backward iteration, until detecting not
To more tiny profile, you can think and do not comprise lump in this frame.
2. dividing method according to claim 1 is it is characterised in that pretreatment image in B, including following two steps
Suddenly：
1) intercept ROI, take out the minimum image rectangular area comprising mammary gland；
2) morphology enhancing is carried out using top cap (top-hat) computing, obtain image after pretreatment.
3. dividing method according to claim 1 is it is characterised in that to image coarse segmentation after pretreatment, determine focus
Initial profile, comprises the following steps：
1) super-pixel segmentation：Super-pixel segmentation is carried out using SLIC clustering algorithm；
2) suspicious region screening, using by the super-pixel region merging technique of close gray value, only retains the higher region of gray value and makees
For the method for suspicious region, screened, obtained the image after coarse segmentation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610761183.8A CN106447682A (en) | 2016-08-29 | 2016-08-29 | Automatic segmentation method for breast MRI focus based on Inter-frame correlation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610761183.8A CN106447682A (en) | 2016-08-29 | 2016-08-29 | Automatic segmentation method for breast MRI focus based on Inter-frame correlation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106447682A true CN106447682A (en) | 2017-02-22 |
Family
ID=58090626
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610761183.8A Pending CN106447682A (en) | 2016-08-29 | 2016-08-29 | Automatic segmentation method for breast MRI focus based on Inter-frame correlation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106447682A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016683A (en) * | 2017-04-07 | 2017-08-04 | 衢州学院 | The level set hippocampus image partition method initialized based on region growing |
| CN107016681A (en) * | 2017-03-29 | 2017-08-04 | 浙江师范大学 | Brain MRI lesion segmentation approach based on full convolutional network |
| CN107610095A (en) * | 2017-08-04 | 2018-01-19 | 南京邮电大学 | Heart CT coronary artery full-automatic partition methods based on image co-registration |
| CN107705290A (en) * | 2017-09-05 | 2018-02-16 | 哈尔滨工业大学深圳研究生院 | A kind of AMOLED display screens Mura defects detection method |
| CN108510470A (en) * | 2018-02-12 | 2018-09-07 | 佛山市顺德区中山大学研究院 | Super-pixel extraction transitional period lamellar corneal festers the method in region |
| CN108629785A (en) * | 2018-05-10 | 2018-10-09 | 西安电子科技大学 | Based on the three-dimensional magnetic resonance pancreas image partition method from step study |
| CN108682015A (en) * | 2018-05-28 | 2018-10-19 | 科大讯飞股份有限公司 | Lesion segmentation method, apparatus, equipment and storage medium in a kind of biometric image |
| CN108846396A (en) * | 2018-05-25 | 2018-11-20 | 广州杰赛科技股份有限公司 | Picture material dividing method, device and licence plate recognition method |
| CN108961278A (en) * | 2018-06-20 | 2018-12-07 | 深圳市旭东数字医学影像技术有限公司 | The method and its system of abdominal wall muscle segmentation based on image data |
| CN109214388A (en) * | 2018-11-21 | 2019-01-15 | 济南浪潮高新科技投资发展有限公司 | A kind of lesion segmentation approach and device based on personalized converged network |
| CN109785293A (en) * | 2018-12-22 | 2019-05-21 | 昆明理工大学 | A method of the ultrasound image lesion based on global and local movable contour model is divided automatically |
| CN110148111A (en) * | 2019-04-01 | 2019-08-20 | 江西比格威医疗科技有限公司 | The automatic testing method of a variety of retina lesions in a kind of retina OCT image |
| WO2020029064A1 (en) * | 2018-08-07 | 2020-02-13 | 温州医科大学 | Optical coherence tomographic image processing method |
| CN110827311A (en) * | 2019-11-05 | 2020-02-21 | 中铁十一局集团电务工程有限公司 | Cable conductor sectional area measuring method and system based on imaging method |
| WO2020077962A1 (en) * | 2018-10-16 | 2020-04-23 | 杭州依图医疗技术有限公司 | Method and device for breast image recognition |
| CN111627005A (en) * | 2020-05-25 | 2020-09-04 | 殷晓峰 | Fracture area identification method and system for bone subdivision based on shape |
| CN111862125A (en) * | 2019-04-28 | 2020-10-30 | 株式会社理光 | Contour segmentation method and device and computer readable storage medium |
| WO2022101768A1 (en) * | 2020-11-16 | 2022-05-19 | International Business Machines Corporation | Medical object detection and identification |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104599270A (en) * | 2015-01-18 | 2015-05-06 | 北京工业大学 | Breast neoplasms ultrasonic image segmentation method based on improved level set algorithm |
| CN104715484A (en) * | 2015-03-20 | 2015-06-17 | 中国科学院自动化研究所 | Automatic tumor area partition method based on improved level set |
| CN104766340A (en) * | 2015-04-30 | 2015-07-08 | 上海联影医疗科技有限公司 | Image segmentation method |
| CN105389823A (en) * | 2015-12-04 | 2016-03-09 | 浙江工业大学 | Interactive intelligent image segmentation method based on tumor attack |
-
2016
- 2016-08-29 CN CN201610761183.8A patent/CN106447682A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104599270A (en) * | 2015-01-18 | 2015-05-06 | 北京工业大学 | Breast neoplasms ultrasonic image segmentation method based on improved level set algorithm |
| CN104715484A (en) * | 2015-03-20 | 2015-06-17 | 中国科学院自动化研究所 | Automatic tumor area partition method based on improved level set |
| CN104766340A (en) * | 2015-04-30 | 2015-07-08 | 上海联影医疗科技有限公司 | Image segmentation method |
| CN105389823A (en) * | 2015-12-04 | 2016-03-09 | 浙江工业大学 | Interactive intelligent image segmentation method based on tumor attack |
Non-Patent Citations (6)
| Title |
|---|
| CHUNMING LI 等: "Distance Regularized Level Set Evolution and Its Application to Image Segmentation", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| JINGHUI CHU 等: "A novel computer aided breast mass detection scheme based on morphological enhancement and SLIC superpixel segmentation", 《MEDICAL PHYSICS》 * |
| RADHAKRISHNA ACHANTA 等: "SLIC Superpixels Compared to State-of-the-Art Superpixel Methods", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| XIAO-FENGWANG 等: "An efficient local Chan-Vese model for image segmentation", 《PATTERN RECOGNITION》 * |
| 朱益苗 等: "基于改进Marching Cubes算法的乳腺MRI肿块三维重建", 《传感技术学报》 * |
| 李艳诚: "基于MRI的脑肿瘤序列图像的分割方法的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016681B (en) * | 2017-03-29 | 2023-08-25 | 浙江师范大学 | Brain MRI tumor segmentation method based on full convolution network |
| CN107016681A (en) * | 2017-03-29 | 2017-08-04 | 浙江师范大学 | Brain MRI lesion segmentation approach based on full convolutional network |
| CN107016683A (en) * | 2017-04-07 | 2017-08-04 | 衢州学院 | The level set hippocampus image partition method initialized based on region growing |
| CN107610095A (en) * | 2017-08-04 | 2018-01-19 | 南京邮电大学 | Heart CT coronary artery full-automatic partition methods based on image co-registration |
| CN107705290A (en) * | 2017-09-05 | 2018-02-16 | 哈尔滨工业大学深圳研究生院 | A kind of AMOLED display screens Mura defects detection method |
| CN107705290B (en) * | 2017-09-05 | 2021-07-02 | 哈尔滨工业大学深圳研究生院 | Mura defect detection method for AMOLED display screen |
| CN108510470A (en) * | 2018-02-12 | 2018-09-07 | 佛山市顺德区中山大学研究院 | Super-pixel extraction transitional period lamellar corneal festers the method in region |
| CN108510470B (en) * | 2018-02-12 | 2020-08-04 | 佛山市顺德区中山大学研究院 | Method for extracting transition-period sheet corneal ulceration area by super pixels |
| CN108629785A (en) * | 2018-05-10 | 2018-10-09 | 西安电子科技大学 | Based on the three-dimensional magnetic resonance pancreas image partition method from step study |
| CN108629785B (en) * | 2018-05-10 | 2021-09-10 | 西安电子科技大学 | Three-dimensional magnetic resonance pancreas image segmentation method based on self-learning |
| CN108846396A (en) * | 2018-05-25 | 2018-11-20 | 广州杰赛科技股份有限公司 | Picture material dividing method, device and licence plate recognition method |
| CN108846396B (en) * | 2018-05-25 | 2020-11-13 | 广州杰赛科技股份有限公司 | Image content segmentation method and device and license plate recognition method |
| CN108682015A (en) * | 2018-05-28 | 2018-10-19 | 科大讯飞股份有限公司 | Lesion segmentation method, apparatus, equipment and storage medium in a kind of biometric image |
| CN108961278A (en) * | 2018-06-20 | 2018-12-07 | 深圳市旭东数字医学影像技术有限公司 | The method and its system of abdominal wall muscle segmentation based on image data |
| CN108961278B (en) * | 2018-06-20 | 2023-01-31 | 深圳市旭东数字医学影像技术有限公司 | Method and system for abdominal wall muscle segmentation based on image data |
| WO2020029064A1 (en) * | 2018-08-07 | 2020-02-13 | 温州医科大学 | Optical coherence tomographic image processing method |
| WO2020077962A1 (en) * | 2018-10-16 | 2020-04-23 | 杭州依图医疗技术有限公司 | Method and device for breast image recognition |
| CN109214388A (en) * | 2018-11-21 | 2019-01-15 | 济南浪潮高新科技投资发展有限公司 | A kind of lesion segmentation approach and device based on personalized converged network |
| CN109785293B (en) * | 2018-12-22 | 2022-09-27 | 昆明理工大学 | Ultrasonic image focus segmentation method based on global and local active contour models |
| CN109785293A (en) * | 2018-12-22 | 2019-05-21 | 昆明理工大学 | A method of the ultrasound image lesion based on global and local movable contour model is divided automatically |
| CN110148111A (en) * | 2019-04-01 | 2019-08-20 | 江西比格威医疗科技有限公司 | The automatic testing method of a variety of retina lesions in a kind of retina OCT image |
| CN111862125A (en) * | 2019-04-28 | 2020-10-30 | 株式会社理光 | Contour segmentation method and device and computer readable storage medium |
| CN111862125B (en) * | 2019-04-28 | 2024-06-21 | 株式会社理光 | Contour segmentation method and device and computer readable storage medium |
| CN110827311A (en) * | 2019-11-05 | 2020-02-21 | 中铁十一局集团电务工程有限公司 | Cable conductor sectional area measuring method and system based on imaging method |
| CN110827311B (en) * | 2019-11-05 | 2023-07-21 | 中铁十一局集团电务工程有限公司 | Imaging method-based cable conductor sectional area measurement method and system |
| CN111627005A (en) * | 2020-05-25 | 2020-09-04 | 殷晓峰 | Fracture area identification method and system for bone subdivision based on shape |
| CN111627005B (en) * | 2020-05-25 | 2023-02-17 | 殷晓峰 | Fracture area identification method and system for bone subdivision based on shape |
| WO2022101768A1 (en) * | 2020-11-16 | 2022-05-19 | International Business Machines Corporation | Medical object detection and identification |
| US11580337B2 (en) | 2020-11-16 | 2023-02-14 | International Business Machines Corporation | Medical object detection and identification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106447682A (en) | Automatic segmentation method for breast MRI focus based on Inter-frame correlation | |
| CN105389811B (en) | A kind of multi-modality medical image processing method split based on multilevel threshold | |
| CN105719324B (en) | Image processing apparatus and image processing method | |
| Acharya et al. | Particle swarm optimized texture based histogram equalization (PSOTHE) for MRI brain image enhancement | |
| WO2019000455A1 (en) | Method and system for segmenting image | |
| CN107886514A (en) | Breast molybdenum target image lump semantic segmentation method based on depth residual error network | |
| CN104268873B (en) | Breast tumor partition method based on nuclear magnetic resonance images | |
| CN107230206A (en) | A kind of 3D Lung neoplasm dividing methods of the super voxel sequence lung images based on multi-modal data | |
| CN108564567A (en) | A kind of ultrahigh resolution pathological image cancerous region method for visualizing | |
| EP1815429A2 (en) | Detecting and classifying lesions in ultrasound images | |
| CN109087703A (en) | Abdominal cavity CT image peritonaeum metastatic marker method based on depth convolutional neural networks | |
| Maitra et al. | Automated digital mammogram segmentation for detection of abnormal masses using binary homogeneity enhancement algorithm | |
| CN112263217B (en) | Improved convolutional neural network-based non-melanoma skin cancer pathological image lesion area detection method | |
| Maitra et al. | Detection of abnormal masses using divide and conquer algorithmin digital mammogram | |
| CN109498046A (en) | The myocardial infarction quantitative evaluating method merged based on nucleic image with CT coronary angiography | |
| CN113139981A (en) | DCE-MRI (direct current imaging-magnetic resonance imaging) breast tumor image segmentation method based on deep neural network | |
| CN112686875A (en) | Tumor prediction method of PET-CT image based on neural network and computer readable storage medium | |
| KR101078756B1 (en) | Method for Detection of Clustered Microcalcifications on Digital Mammogram | |
| CN109919216B (en) | Counterlearning method for computer-aided diagnosis of prostate cancer | |
| CN105869169B (en) | A kind of automatic division method of the micro- arrangement image of tumor tissues | |
| CN110458842A (en) | Brain tumor dividing method based on the three-dimensional intensive connection network of binary channels | |
| CN112348826B (en) | Interactive liver segmentation method based on geodesic distance and V-net | |
| CN104331864B (en) | Based on the processing of the breast image of non-down sampling contourlet and the significant model of vision | |
| Tung et al. | Identifying pathological slices of gastric cancer via deep learning | |
| CN112967254A (en) | Lung disease identification and detection method based on chest CT image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170222 |
|
| RJ01 | Rejection of invention patent application after publication |