CN106023239A - Breast lump segmentation system and method based on mammary gland subarea density clustering - Google Patents

Breast lump segmentation system and method based on mammary gland subarea density clustering Download PDF

Info

Publication number
CN106023239A
CN106023239A CN201610523605.8A CN201610523605A CN106023239A CN 106023239 A CN106023239 A CN 106023239A CN 201610523605 A CN201610523605 A CN 201610523605A CN 106023239 A CN106023239 A CN 106023239A
Authority
CN
China
Prior art keywords
mammary gland
subregion
image
gray
feature
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
Application number
CN201610523605.8A
Other languages
Chinese (zh)
Inventor
信俊昌
李默
田硕
马春晓
高铭泽
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.)
Northeastern University China
Original Assignee
Northeastern University China
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 Northeastern University China filed Critical Northeastern University China
Priority to CN201610523605.8A priority Critical patent/CN106023239A/en
Publication of CN106023239A publication Critical patent/CN106023239A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • 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
    • 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
    • G06T2207/30096—Tumor; Lesion

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

The invention provides a breast lump segmentation system and method based on mammary gland subarea density clustering, and relates to the technical field of medical image post processing. After early-stage image preprocessing is performed on a mammary gland image through a preprocessing unit, a segmentation window unit segments the image into a plurality of subareas, density feature extraction is performed on each subarea by use of a density feature extraction unit, a cluster unit performs clustering, and finally, a segmentation result visual unit displays the image after clustering segmentation. The breast lump segmentation system and method based on mammary gland subarea density clustering, provided by the invention, employ a method based on mammary gland subarea density clustering, carry out accurate segmentation of breast lumps, clearly display lump positions after segmentation, and can effectively provide assistance in accurate diagnosis of breast diseases.

Description

Mammary gland tumor segmenting system based on mammary gland subregion Density Clustering and method
Technical field:
The present invention relates to medical image post-procession technique field, particularly relate to a kind of based on mammary gland subregion Density Clustering Mammary gland tumor segmenting system and method.
Background technology:
Breast carcinoma is the modal malignancy disease of women, and early discovery, early diagnosis, early treatment are to reduce mammary gland The key of cancer hazardness.The doubtful lump with breast carcinoma characteristic can be split, detect by computer-aided diagnosis technology And classification.Wherein, the segmentation of doubtful lump is the basis of Computer-aided Diagnosis of Breast Cancer.Meanwhile, breast density is breast cancer bump Key character.The research of lump doubtful to mammary gland the most both at home and abroad has a lot, is roughly divided into two classes, and a class is side based on edge Method, an other class is method based on region, but this two classes method all can not well embody the density of mammary gland tumor, and this is heavy Want feature.
Summary of the invention:
For the defect of prior art, the present invention provides a kind of mammary gland tumor based on mammary gland subregion Density Clustering to split System and method, uses method based on mammary gland subregion Density Clustering, carries out the accurate segmentation of mammary gland tumor, can effectively assist The Accurate Diagnosis of mastopathy.
On the one hand, the present invention provides a kind of mammary gland tumor segmenting system based on mammary gland subregion Density Clustering, this system Including: pretreatment unit, split window unit, density feature extraction unit, cluster cell and segmentation result visualization.
Pretreatment unit includes that image denoising device, image intensifier, image gray levels variator and mammary gland edge coordinate are raw Grow up to be a useful person.
Image denoising device, for galactophore image is carried out denoising, obtains the image after denoising；
Image intensifier, overall or local characteristics, difference portion in expanded view picture in the galactophore image after emphasizing denoising The difference of interdigit, suppression are lost interest in the expression in region and increase the contrast of doubtful lump and surrounding tissue, obtain enhancing figure Picture；
Image gray levels variator, for being compressed the gray level strengthening image, in order to ensuing feature extraction Unit uses；
Mammary gland edge coordinate maker, for obtaining the edge coordinate of mammary gland part, lump is extracted by shielding background area Impact.
Split window unit includes sliding window generator and subregion dispenser.
Sliding window generator, for generating the upper left corner being placed on mammary gland part in the mammary gland CC image of left side or the right side The square sliding window in the upper right corner of mammary gland part in the mammary gland CC image of side；
Subregion dispenser, for mammary gland part in galactophore image is divided into several overlapped subregions, makees The basis of post-processing unit for it.
Density feature extraction unit include subregion histogrammic average extractor, variance extractor, degree of skewness extractor, Peak extractor, gray-scale intensity average extractor, gray-scale intensity variance extractor, gray-scale intensity degree of skewness extractor and gray scale are close Degree peak extractor, is respectively used to extract the character pair value of each described subregion.
Cluster cell is for using clustering algorithm to cluster the character pair value of described subregion.
Segmentation result visualization includes lump segmentation result display and mass edge coordinate generator.
Lump segmentation result display, for showing the result visualization of subregion cluster, in order to Ke Yiqing Clear sees place, lump position；
Mass edge coordinate generator, for extracting the edge coordinate of lump, in order to test accuracy rate, and is that breast carcinoma is auxiliary Diagnosis is helped to provide basis.
Further, average extractor, for extracting the average of all subregion, it is used for describing the mean density value of subregion, Mean value computation formula is:
v 1 = Σ i = 0 L - 1 z i p ( z i )
Wherein, v1For the mean density value of subregion, L is the gray level of image, ziIt is the pixel count of i for gray scale, p (zi) it is Gray scale is the ratio that the pixel count of i accounts for all pixel counts；
Variance extractor, for extracting the variance of all subregion, is used for describing the situation of the variable density of subregion, variance Computing formula is:
v 2 = Σ i = 0 L - 1 ( z i - m ) 2 p ( z i )
Wherein, v2For the density variance of subregion, m is the gray average of image；
Degree of skewness extractor, for extracting the oblique degree of bias of all subregion, is used for describing the symmetry of subregion Density Distribution, Degree of skewness computing formula is:
v 3 = 1 v 1 3 / 2 Σ i = 0 L - 1 ( z i - m ) 3 p ( z i )
Wherein, v3The oblique degree of bias of density for subregion；
Peak extractor, for extracting the peak value of all subregion, is used for describing the peak value of subregion density, and peak computational is public Formula is:
v 4 = 1 v 1 2 Σ i = 0 L - 1 ( z i - m ) 4 p ( z i )
Wherein, v4Density peaks for subregion；
Gray-scale intensity variance extractor, for extracting the gray-scale intensity variance of all subregion, is used for describing the ash of subregion The situation of change of degree density, gray-scale intensity variance computing formula is:
v 5 = Σ i = 0 L - 1 ( z i - p m ) 2 p ( z i )
Wherein, v5For the gray-scale intensity variance of subregion, pmFor the gray-scale intensity average of subregion,
Gray-scale intensity degree of skewness extractor, for extracting the oblique degree of bias of gray-scale intensity of all subregion, is used for describing subregion Gray-scale intensity indexing symmetry, the computing formula of gray-scale intensity degree of skewness is:
v 6 = 1 v 1 3 / 2 Σ i = 0 L - 1 ( z i - p m ) 3 p ( z i )
Wherein, v6The oblique degree of bias of gray-scale intensity for subregion；
Gray-scale intensity peak extractor, for extracting the gray-scale intensity peak value of all subregion, is used for describing the ash of subregion Degree density peaks, the computing formula of gray-scale intensity peak value is:
v 7 = 1 v 1 2 Σ i = 0 L - 1 ( z i - p m ) 4 p ( z i )
Wherein, v7Gray-scale intensity peak value for subregion.
Further, cluster cell includes transformation matrix maker, random parameter maker, transducer, dimensionality reduction device and gathers Class device；
Transformation matrix maker, for according to unsupervised ELM algorithm (US-ELM) principle, generates breast molybdenum target image each The Laplace transform matrix of sub regions eigenmatrix；
Random parameter maker, for according to set hidden node number, the weight vectors of stochastic generation input node and The threshold value of hidden node；
Transducer, for according to unsupervised ELM algorithm principle, utilizes the weight vectors of input node and hidden node Threshold value generates the hidden layer output matrix of the mammary gland subregion feature in US-ELM；
Dimensionality reduction device, for according to unsupervised ELM algorithm principle, the Laplce's square generated according to transformation matrix maker Battle array and the hidden layer output matrix of mammary gland subregion feature, obtain the three-dimensional feature after mammary gland subregion Feature Dimension Reduction；
Cluster device, clusters for the three-dimensional feature using clustering algorithm to obtain dimensionality reduction device.
On the other hand, the present invention also provides for a kind of mammary gland tumor dividing method based on mammary gland subregion Density Clustering, should Method comprises the following steps:
Step 1: galactophore image is gone dry, enhancing and the pretreatment of gray-scale compression, and obtains mammary gland edge coordinate；
Step 2: generate sliding window, and whole mammary gland part is split；
Step 3: obtain the histogram feature square variable of each mammary gland subregion；Feature specifically includes: the sub-district of each mammary gland The characteristics of mean in territory, Variance feature, degree of skewness feature, sharp peaks characteristic, gray-scale intensity Variance feature, gray-scale intensity degree of skewness feature With gray-scale intensity sharp peaks characteristic, each described feature is column matrix；
Step 4: feature step 3 obtained clusters；
Step 5: show visual for the mammary gland tumor segmentation result obtained after cluster, and the image of display is entered Row rim detection, obtains the edge coordinate of lump.
Further, step 1 specifically includes:
Step 101: galactophore image is carried out mean filter operation, obtains the image after denoising；
Step 102: the image after denoising carries out contrast and strengthens operation, obtains strengthening image；
Step 103: enhancing image is carried out gray-scale compression, the image after being compressed；
Step 104: carry out edge detecting operation to strengthening image, it is thus achieved that mammary gland edge coordinate.
Further, step 2 specifically includes:
Step 201: generate a square sliding window, and utilize the mammary gland edge coordinate that step 1 obtains, allows this window It is placed on the upper right corner of mammary gland part in the upper left corner of mammary gland part in the mammary gland CC image of left side or right side breast CC image；
Step 202: utilize the square sliding window generated, with pre-fixed step size, start downwards from the starting point coordinate of mammary gland Sliding, when window top left co-ordinate is beyond mammary gland scope, sliding window returns to mammary gland top, and a step-length of sliding to the right Distance, slides the most still further below, circulates successively, until sliding window leaves mammary region, i.e. obtains several and mutually hands over The mammary gland subregion of fork.
Further, in step 4, the feature utilizing unsupervised ELM algorithm and clustering algorithm step 3 to be obtained is gathered Class, specifically includes:
Step 401: according to the principle of US-ELM, generate the Laplacian Matrix of input node；
Step 402: set hidden node number, and the weight vectors of stochastic generation input node and the threshold of hidden node Value；
Step 403: according to the principle of US-ELM, utilizes the weight vectors of input node and the threshold value of hidden node, will step The hidden layer output square of the mammary gland subregion feature that the eigenmatrix that the feature square vector that rapid 3 obtain is constituted is converted in US-ELM Battle array；
Step 404: according to the principle of US-ELM, utilizes Laplacian Matrix that step 401 generates and step 403 to obtain The hidden layer output matrix of mammary gland subregion feature, obtains the three-dimensional feature matrix after mammary gland subregion eigenmatrix dimensionality reduction；
Step 405: utilize ambiguity of space angle C means clustering algorithm three-dimensional feature to be clustered, by three-dimensional feature matrix Often row regards a point as, is divided into 3 classes, respectively C1、C2And C3。
Further, in steps of 5, the visual detailed process shown of mammary gland tumor segmentation result that will obtain For: utilize a blank image, the point being divided into a class is shown on blank image, obtain sorted after cluster Image, displays it and i.e. can get visual mammary gland tumor segmentation result.
Further, square sliding window pixel size is 32 × 32.
Further, the gray-scale compression of image is strengthened by 12 boil down tos 8.
As shown from the above technical solution, mammary gland tumor based on the mammary gland subregion Density Clustering segmentation that the present invention provides is System and method, after galactophore image is carried out early stage Image semantic classification, use method based on mammary gland subregion Density Clustering, carry out The accurate segmentation of mammary gland tumor, and clearly indicate out by lump position after segmentation, can effectively assist the accurate of mastopathy Diagnosis.
Accompanying drawing illustrates:
Fig. 1 is mammograms pending in the embodiment of the present invention；
The system architecture diagram that Fig. 2 provides for the embodiment of the present invention；
The method flow diagram that Fig. 3 provides for the embodiment of the present invention；
Fig. 4 is the flow chart of steps in Fig. 3 clustered eigenmatrix；
Fig. 5 is the image in the embodiment of the present invention after mammary gland tumor segmentation.
In figure: 1, pretreatment unit；2, split window unit；3, density feature extraction unit；4, cluster cell；5, segmentation Result visualization unit.
Detailed description of the invention:
Below in conjunction with the accompanying drawings and embodiment, the detailed description of the invention of the present invention is described in further detail.Hereinafter implement Example is used for illustrating the present invention, but is not limited to the scope of the present invention.
In the present embodiment, pending mammograms is as it is shown in figure 1, this original galactophore image is (I1, I2..., IN).Fig. 2 shows the structured flowchart of lump segmenting system based on mammary gland subregion Density Clustering, as in figure 2 it is shown, this enforcement The system of example includes: pretreatment unit 1, split window unit 2, density feature extraction unit 3, cluster cell 4 and segmentation result Visualization 5.
Pretreatment unit 1 includes that image denoising device, image intensifier, image gray levels variator and mammary gland edge coordinate are raw Grow up to be a useful person.
Image denoising device, for galactophore image (I1, I2..., IN) carry out denoising operation, reduce mammograms In noise, obtain the image (U after denoising1, U2..., UN)。
Image intensifier, the galactophore image (U after emphasizing denoising1, U2..., UNIn), overall or local characteristics, expands Difference between different parts in image, suppression is lost interest in the expression in region, increases the contrast of doubtful lump and surrounding tissue, Obtain strengthening image (T1, T2..., TN)。
Image gray levels variator, for strengthening image (T1, T2..., TN) gray level by 12 boil down tos 8, To compression image (G1, G2..., GN), in order to ensuing feature extraction unit uses.
Mammary gland edge coordinate maker, for obtaining the edge coordinate of mammary gland part, lump is extracted by shielding background area Impact.
Split window unit 2 includes sliding window generator and subregion dispenser.
Sliding window generator, for generating the square window of 32 × 32, and allow this window be placed on through Pretreatment unit 1 process after left side mammary gland CC image in mammary gland part in the upper left corner of mammary gland part or right side breast CC image The upper right corner.
Subregion dispenser, for being divided into several overlapped subregion (S by mammary gland part in galactophore image1, S2..., Sq), as the basis of post-processing unit, wherein, q is the number of subregion.
Density feature extraction unit 3 includes that subregion histogrammic average extractor, variance extractor, degree of skewness are extracted Device, peak extractor, gray-scale intensity variance extractor, gray-scale intensity degree of skewness extractor and gray-scale intensity peak extractor.
Average extractor, for extracting the average of all subregion, is used for describing the mean density value of subregion, and mean value computation is public Formula is:
v 1 = Σ i = 0 L - 1 z i p ( z i )
Wherein, v1For the mean density value of subregion, L is the gray level of image, ziIt is the pixel count of i for gray scale, p (zi) it is Gray scale is the ratio that the pixel count of i accounts for all pixel counts；
Variance extractor, for extracting the variance of all subregion, is used for describing the situation of the variable density of subregion, variance Computing formula is:
v 2 = Σ i = 0 L - 1 ( z i - m ) 2 p ( z i )
Wherein, v2For the density variance of subregion, m is the gray average of image；
Degree of skewness extractor, for extracting the oblique degree of bias of all subregion, is used for describing the symmetry of subregion Density Distribution, Degree of skewness computing formula is:
v 3 = 1 v 1 3 / 2 Σ i = 0 L - 1 ( z i - m ) 3 p ( z i )
Wherein, v3The oblique degree of bias of density for subregion；
Peak extractor, for extracting the peak value of all subregion, is used for describing the peak value of subregion density, and peak computational is public Formula is:
v 4 = 1 v 1 2 Σ i = 0 L - 1 ( z i - m ) 4 p ( z i )
Wherein, v4Density peaks for subregion；
Gray-scale intensity variance extractor, for extracting the gray-scale intensity variance of all subregion, is used for describing the ash of subregion The situation of change of degree density, gray-scale intensity variance computing formula is:
v 5 = Σ i = 0 L - 1 ( z i - p m ) 2 p ( z i )
Wherein, v5For the gray-scale intensity variance of subregion, pmFor the gray-scale intensity average of subregion,
Gray-scale intensity degree of skewness extractor, for extracting the oblique degree of bias of gray-scale intensity of all subregion, is used for describing subregion Gray-scale intensity indexing symmetry, the computing formula of gray-scale intensity degree of skewness is:
v 6 = 1 v 1 3 / 2 Σ i = 0 L - 1 ( z i - p m ) 3 p ( z i )
Wherein, v6The oblique degree of bias of gray-scale intensity for subregion；
Gray-scale intensity peak extractor, for extracting the gray-scale intensity peak value of all subregion, is used for describing the ash of subregion Degree density peaks, the computing formula of gray-scale intensity peak value is:
v 7 = 1 v 1 2 Σ i = 0 L - 1 ( z i - p m ) 4 p ( z i )
Wherein, v7Gray-scale intensity peak value for subregion.
Cluster cell 4 includes transformation matrix maker, random parameter maker, transducer, dimensionality reduction device and cluster device.
Transformation matrix maker, according to unsupervised ELM (referred to as US-ELM) principle, is used for generating breast molybdenum target image 7 × q dimensional feature matrix V of each sub regions1、V2、…、V7Laplace transform matrix L, wherein, Vj(j=1,2 ..., 7) It is the column vector of 1 × q dimension, by the eigenvalue v of density feature extraction unit calculated q sub regionsjComposition.
Random parameter maker, sets hidden node number s, the weight vectors ω of stochastic generation input node1、ω2、…、 ωs-1、ωs, and threshold value b of hidden node1、b2、…、bs-1、bs。
Transducer, according to US-ELM principle, is used for the weight vectors of the input node utilizing random parameter maker to obtain ω1、ω2、…、ωs-1、ωs, and threshold value b of hidden node1、b2、…、bs-1、bsGenerate the mammary gland subregion feature in US-ELM Hidden layer output matrix H0；
Dimensionality reduction device, according to US-ELM principle, for the Laplacian Matrix L generated according to transformation matrix maker and mammary gland The hidden layer output matrix H of subregion feature0, obtain the three-dimensional feature F after mammary gland subregion Feature Dimension Reduction1、F2And F3。
Cluster device, uses the three-dimensional feature F that dimensionality reduction device is obtained by clustering algorithm1、F2And F3Cluster.
Segmentation result visualization 5 includes lump segmentation result display and mass edge coordinate generator.
Lump segmentation result display, for showing the result visualization of subregion cluster, in order to Ke Yiqing Clear sees place, lump position.
Mass edge coordinate generator, for extracting the edge coordinate of lump, in order to test accuracy rate, and is that breast carcinoma is auxiliary Diagnosis is helped to provide basis.
The present embodiment also provides for a kind of segmentation system utilizing above-mentioned mammary gland tumor based on mammary gland subregion Density Clustering The dividing method of the mammary gland tumor that system realizes, as it is shown on figure 3, specifically include following steps.
Step 1: carry out galactophore image pretreatment, carries out dry, enhancing and gray-scale compression, and obtains mammary gland edge seat Mark, specifically includes:
Step 101: utilize image denoising device to galactophore image (I1, I2..., IN) carry out mean filter operation, obtain denoising After image (U1, U2..., UN)；
Step 102: utilize image intensifier to the image (U after denoising1, U2..., UN) carry out contrast enhancing operation, Image (T after enhancing1, T2..., TN)；
Step 103: image gray levels variator is by enhanced image (T1, T2..., TN) carry out gray-scale compression, from 12 Position becomes 8, obtains compressing image (G1, G2..., GN)；
Step 104: mammary gland edge coordinate maker is to enhanced image (G1, G2..., GN) carry out edge detecting operation, Obtain mammary gland edge coordinateWherein, M is the number of edge coordinate.
Step 2: generate sliding window, and whole mammary gland part is split, specifically include:
Step 201: utilize the square sliding window that sliding window generator will generate 32 × 32, and utilize step The mammary gland edge coordinate that 1 obtainsThis window is allowed to be placed on the upper left corner or the right side of mammary gland part in the mammary gland CC image of left side The upper right corner of mammary gland part in the mammary gland CC image of side；
Step 202: utilize subregion dispenser by square sliding window with 16 as step-length, open from the starting point coordinate of mammary gland Beginning slide downward, when window top left co-ordinate is beyond mammary gland scope, sliding window returns to mammary gland top, and slides one to the right The distance of step-length, slides the most still further below, circulates successively, until sliding window leaves mammary region, has i.e. obtained q and has done Mammary gland subregion (the S of the individual overlap that intersects1, S2..., Sq)。
Step 3: obtain the histogram feature matrix V of each mammary gland subregion1、V2、…、V7, specifically include:
Step 301: utilize average extractor to obtain the characteristics of mean of each mammary gland subregion, constitutes 1 × q and ties up matrix V1；
Step 302: utilize variance extractor to obtain the Variance feature of each mammary gland subregion, constitutes l × q and ties up matrix V2；
Step 303: utilize degree of skewness extractor to obtain the degree of skewness feature of each mammary gland subregion, constitutes l × q and ties up matrix V3；
Step 304: utilize peak extractor to obtain the sharp peaks characteristic of each mammary gland subregion, constitutes l × q and ties up matrix V4；
Step 305: utilize gray-scale intensity variance extractor to obtain the gray-scale intensity Variance feature of each mammary gland subregion, structure 1 × q is become to tie up matrix V5；
Step 306: utilize gray-scale intensity degree of skewness extractor to obtain the gray-scale intensity degree of skewness spy of each mammary gland subregion Levy, constitute 1 × q and tie up matrix V6；
Step 307: utilize gray-scale intensity peak extractor to obtain the gray-scale intensity sharp peaks characteristic of each mammary gland subregion, structure 1 × q is become to tie up matrix V7。
The density feature matrix obtained in the present embodiment is:
V1=(208.796,53.519 ..., 1024)；
V2=(80568.0,5663.0 ..., 1048576)；
V3=(119776.0,65795.0 ..., 41285)；
V4=(151317.0,295337.0 ..., 41285)；
V5=(0.0007843,0.0001896 ..., 0.0039)；
V6=(0.03734,0.0101 ..., 0.1563)；
V7=(221.5038,189.2406 ..., 252.9961).
Step 4: the feature utilizing US-ELM algorithm and clustering algorithm step 3 to be obtained clusters, as shown in Figure 4, for Cluster flow chart, specifically includes:
Step 401: in transformation matrix maker, utilizes the principle of US-ELM, generates Laplce's square of input node Battle array L；
Step 402: in random parameter maker, sets hidden node number, and the weight of stochastic generation input node Vector ω1、ω2、…、ωs-1、ωs, and threshold value b of hidden node1、b2、…、bs-1、bs；
The weight vectors of the input node obtained in the present embodiment is:
ω1=(-0.6233,0.837l ..., 0.2845)；
ω2=(0.2324,0.695l ..., 0.8341)；
…
ω5939=(0.8637,0.2452 ..., 0.6376)；
The hidden node threshold value obtained is:
b1=0.2344；
b2=0.2344；
…
b5939=0.2344；
Step 403: in the converter, according to the principle of US-ELM, utilizes the input node that random parameter maker obtains Weight vectors ω1、ω2、…、ωs-1、ωs, and threshold value b of hidden node1、b2、…、bs-1、bs, by breast molybdenum target image 7 × q dimensional feature matrix V of subregion1、V2、…、V7The hidden layer output matrix of the mammary gland subregion feature being converted in US-ELM H0；
Step 404: in dimensionality reduction device, according to the principle of US-ELM, utilizes the Laplce that transformation matrix maker generates The hidden layer output matrix H of matrix L and mammary gland subregion feature0, obtain the three-dimensional feature F after mammary gland subregion Feature Dimension Reduction1、F2 And F3；
The present embodiment obtains the three-dimensional feature F after mammary gland subregion density feature dimensionality reduction1、F2And F3It is respectively as follows:
F1=(0.0158,0.0184 ..., 0.0073)；
F2=(-0.0428 ,-0.0363 ..., 0.0017)；
F3=(0.0062,0.0384 ..., 0.0010)；
Step 405: in cluster device, utilize ambiguity of space angle C means clustering algorithm to three-dimensional feature F1、F2And F3Gather Class, regards matrix often row as a point, is divided into 3 classes, respectively C1、C2And C3。
Step 5: obtain mammary gland tumor segmentation visualization result and mammary gland tumor edge coordinate, specifically include:
Step 501: according to cluster result, be marked breast molybdenum target image, lump segmentation result display is by lump Segmentation result visual show.Utilize a blank image (A1, A2..., AN), the point of a class will be divided at this Open and show on image, obtain sorted image (B1, B2..., BN), display it and i.e. can get visual knot Really；
Step 502: the image (B obtained according to lump segmentation result display1, B2..., BN), it is carried out rim detection Operation, i.e. can get the edge coordinate of lump, it is achieved the accurate segmentation of mammary gland tumor, the image after segmentation is as shown in Figure 5.
The mammary gland tumor segmenting system based on mammary gland subregion Density Clustering of present invention offer and method, to galactophore image After carrying out early stage Image semantic classification, use method based on mammary gland subregion Density Clustering, carry out the accurate segmentation of mammary gland tumor, And clearly indicate out by lump position after segmentation, can effectively assist the Accurate Diagnosis of mastopathy.
Last it is noted that various embodiments above is only in order to illustrate technical scheme, it is not intended to limit；To the greatest extent The present invention has been described in detail by pipe with reference to foregoing embodiments, it will be understood by those within the art that: it depends on So the technical scheme described in foregoing embodiments can be modified, or the most some or all of technical characteristic is entered Row equivalent；And these amendments or replacement, do not make the essence of appropriate technical solution depart from the claims in the present invention and limited Fixed scope.

Claims (10)

1. a mammary gland tumor segmenting system based on mammary gland subregion Density Clustering, it is characterised in that this system includes: locate in advance Reason unit (1), split window unit (2), density feature extraction unit (3), cluster cell (4) and segmentation result visualization (5)；
Described pretreatment unit (1) includes image denoising device, image intensifier, image gray levels variator and mammary gland edge coordinate Maker；
Described image denoising device, for galactophore image is carried out denoising, obtains the galactophore image after denoising；
Described image intensifier, overall or local characteristics in the galactophore image after emphasizing described denoising, in expanded view picture not Lose interest in the expression in region and increase the contrast of doubtful lump and surrounding tissue with the difference between position, suppression, increased Strong image；
Described image gray levels variator, for being compressed the gray level of described enhancing image；
Described mammary gland edge coordinate maker, for obtaining the edge coordinate of mammary gland part in galactophore image；
Described split window unit (2) includes sliding window generator and subregion dispenser；
Described sliding window generator, for generating the upper left corner being placed on mammary gland part in the mammary gland CC image of left side or the right side The square sliding window in the upper right corner of mammary gland part in the mammary gland CC image of side；
Described subregion dispenser, for being divided into several overlapped subregions by mammary gland part in galactophore image；
Described density feature extraction unit (3) includes that subregion histogrammic average extractor, variance extractor, degree of skewness are extracted Device, peak extractor, gray-scale intensity variance extractor, gray-scale intensity degree of skewness extractor and gray-scale intensity peak extractor, point Yong Yu not extract the character pair value of each described subregion；
Described cluster cell (4) is for using clustering algorithm to cluster the character pair value of described subregion；
Described segmentation result visualization (5) includes lump segmentation result display and mass edge coordinate generator；
Described lump segmentation result display, for showing the result visualization of described cluster cell (4)；
Described mass edge coordinate generator, for extracting the edge coordinate of lump.
Mammary gland tumor segmenting system based on mammary gland subregion Density Clustering the most according to claim 1, it is characterised in that Described average extractor, for extracting the average of described all subregion, is used for describing the mean density value of described subregion, described all The computing formula of value is:
v 1 = Σ i = 0 L - 1 z i p ( z i )
Wherein, v1For the mean density value of subregion, L is the gray level of image, ziIt is the pixel count of i for gray scale, p (zi) it is gray scale Pixel count for i accounts for the ratio of all pixel counts；
Described variance extractor, for extracting the variance of described all subregion, for describing the variable density of described subregion Situation, the computing formula of described variance is:
v 2 = Σ i = 0 L - 1 ( z i - m ) 2 p ( z i )
Wherein, v2For the density variance of subregion, m is the gray average of image；
Described degree of skewness extractor, for extracting the oblique degree of bias of described all subregion, is used for describing described subregion Density Distribution Symmetry, the computing formula of described degree of skewness is:
v 3 = 1 v 1 3 / 2 Σ i = 0 L - 1 ( z i - m ) 3 p ( z i )
Wherein, v3The oblique degree of bias of density for subregion；
Described peak extractor, for extracting the peak value of described all subregion, is used for describing the peak value of described subregion density, institute The computing formula stating peak value is:
v 4 = 1 v 1 2 Σ i = 0 L - 1 ( z i - m ) 4 p ( z i )
Wherein, v4Density peaks for subregion；
Described gray-scale intensity variance extractor, for extracting the gray-scale intensity variance of described all subregion, is used for describing described son The situation of change of the gray-scale intensity in region, the computing formula of described gray-scale intensity variance is:
v 5 = Σ i = 0 L - 1 2 ( z i - p m ) 2 p ( z i )
Wherein, v5For the gray-scale intensity variance of subregion, pmFor the gray-scale intensity average of subregion,
Described gray-scale intensity degree of skewness extractor, for extracting the oblique degree of bias of gray-scale intensity of described all subregion, is used for describing institute Stating the symmetry of the gray-scale intensity indexing of subregion, the computing formula of described gray-scale intensity degree of skewness is:
v 6 = 1 v 1 3 / 2 Σ i = 0 L - 1 ( z i - p m ) 3 p ( z i )
Wherein, v6The oblique degree of bias of gray-scale intensity for subregion；
Described gray-scale intensity peak extractor, for extracting the gray-scale intensity peak value of described all subregion, is used for describing described son The gray-scale intensity peak value in region, the computing formula of described gray-scale intensity peak value is:
v 7 = 1 v 1 2 Σ i = 0 L - 1 ( z i - p m ) 4 p ( z i )
Wherein, v7Gray-scale intensity peak value for subregion.
Mammary gland tumor segmenting system based on mammary gland subregion Density Clustering the most according to claim 1, it is characterised in that Described cluster cell (4) includes transformation matrix maker, random parameter maker, transducer, dimensionality reduction device and cluster device；
Described transformation matrix maker, for according to unsupervised ELM algorithm (US-ELM) principle, generates breast molybdenum target image each The Laplace transform matrix of sub regions eigenmatrix；
Described random parameter maker, for according to set hidden node number, the weight vectors of stochastic generation input node and The threshold value of hidden node；
Described transducer, for according to unsupervised ELM algorithm principle, utilizing weight vectors and the hidden layer joint of described input node The threshold value of point generates the hidden layer output matrix of the mammary gland subregion feature in US-ELM；
Described dimensionality reduction device, for according to unsupervised ELM algorithm principle, the La Pula generated according to described transformation matrix maker This matrix and the hidden layer output matrix of described mammary gland subregion feature, obtain the three-dimensional feature after mammary gland subregion Feature Dimension Reduction；
Described cluster device, clusters for the three-dimensional feature using clustering algorithm to obtain described dimensionality reduction device.
4. a mammary gland tumor dividing method based on mammary gland subregion Density Clustering, it is characterised in that the method includes following Step:
Step 1: galactophore image is gone dry, enhancing and the pretreatment of gray-scale compression, and obtains mammary gland edge coordinate；
Step 2: generate sliding window, and whole mammary gland part is split；
Step 3: obtain the histogram feature square variable of each mammary gland subregion；Described feature specifically includes: the sub-district of each mammary gland The characteristics of mean in territory, Variance feature, degree of skewness feature, sharp peaks characteristic, gray-scale intensity Variance feature, gray-scale intensity degree of skewness feature With gray-scale intensity sharp peaks characteristic, each described feature is column matrix；
Step 4: the feature that described step 3 obtains is clustered；
Step 5: show visual for the mammary gland tumor segmentation result obtained after cluster, and the image of display is carried out limit Edge detects, and obtains the edge coordinate of lump.
Mammary gland tumor dividing method based on mammary gland subregion Density Clustering the most according to claim 4, it is characterised in that Described step 1 specifically includes:
Step 101: galactophore image is carried out mean filter operation, obtains the image after denoising；
Step 102: the image after described denoising carries out contrast and strengthens operation, obtains strengthening image；
Step 103: described enhancing image is carried out gray-scale compression, the image after being compressed；
Step 104: described enhancing image is carried out edge detecting operation, it is thus achieved that mammary gland edge coordinate.
Mammary gland tumor dividing method based on mammary gland subregion Density Clustering the most according to claim 4, it is characterised in that Described step 2 specifically includes:
Step 201: generate a square sliding window, and utilize the mammary gland edge coordinate that described step 1 obtains, allows this window It is placed on the upper right corner of mammary gland part in the upper left corner of mammary gland part in the mammary gland CC image of left side or right side breast CC image；
Step 202: utilize the described square sliding window generated, with pre-fixed step size, start downwards from the starting point coordinate of mammary gland Sliding, when window top left co-ordinate is beyond mammary gland scope, sliding window returns to mammary gland top, and a step-length of sliding to the right Distance, slides the most still further below, circulates successively, until sliding window leaves mammary region, i.e. obtains several and mutually hands over The mammary gland subregion of fork.
Mammary gland tumor dividing method based on mammary gland subregion Density Clustering the most according to claim 4, it is characterised in that In described step 4, utilize unsupervised ELM algorithm and clustering algorithm the feature that described step 3 obtains to be clustered, specifically wrap Include:
Step 401: according to the principle of US-ELM, generate the Laplacian Matrix of input node；
Step 402: set hidden node number, and the weight vectors of stochastic generation input node and the threshold value of hidden node；
Step 403: according to the principle of US-ELM, utilizes the weight vectors of described input node and the threshold value of hidden node, by institute State the hidden layer output of the mammary gland subregion feature that the eigenmatrix that the feature square vector that step 3 obtains constitutes is converted in US-ELM Matrix；
Step 404: according to the principle of US-ELM, utilizes Laplacian Matrix that described step 401 generates and described step 403 The hidden layer output matrix of the mammary gland subregion feature arrived, obtains the three-dimensional feature matrix after mammary gland subregion eigenmatrix dimensionality reduction；
Step 405: utilize ambiguity of space angle C means clustering algorithm that described three-dimensional feature is clustered, by described three-dimensional feature square The often row of battle array regards a point as, is divided into 3 classes, respectively C1、C2And C3。
8., according to the mammary gland tumor dividing method based on mammary gland subregion Density Clustering described in claim 4 or 7, its feature exists In, in described step 5, described by visual for the mammary gland tumor segmentation result the obtained detailed process shown it is: profit With a blank image, the point being divided into a class is shown on the image of described blank, obtain sorted after cluster Image, displays it and i.e. can get visual mammary gland tumor segmentation result.
9. according to the mammary gland tumor segmenting system based on mammary gland subregion Density Clustering described in any one of claim 1 to 7, its Being characterised by, described square sliding window pixel size is 32 × 32.
10. according to the mammary gland tumor segmenting system based on mammary gland subregion Density Clustering described in any one of claim 1 to 7, It is characterized in that, the gray-scale compression of described enhancing image is by 12 boil down tos 8.
CN201610523605.8A 2016-07-05 2016-07-05 Breast lump segmentation system and method based on mammary gland subarea density clustering Pending CN106023239A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610523605.8A CN106023239A (en) 2016-07-05 2016-07-05 Breast lump segmentation system and method based on mammary gland subarea density clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610523605.8A CN106023239A (en) 2016-07-05 2016-07-05 Breast lump segmentation system and method based on mammary gland subarea density clustering

Publications (1)

Publication Number Publication Date
CN106023239A true CN106023239A (en) 2016-10-12

Family

ID=57107594

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610523605.8A Pending CN106023239A (en) 2016-07-05 2016-07-05 Breast lump segmentation system and method based on mammary gland subarea density clustering

Country Status (1)

Country Link
CN (1) CN106023239A (en)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106780451A (en) * 2016-12-07 2017-05-31 西安电子科技大学 X-ray, ultrasound, infrared image fusion breast lump image detecting method
CN107358267A (en) * 2017-07-20 2017-11-17 东北大学 A kind of breast ultrasound image multivariate classification system and method based on cross-correlation feature
CN107392928A (en) * 2017-07-20 2017-11-24 东北大学 Breast cancer diagnosis system and method based on MRF Snake segmentations
WO2018129650A1 (en) * 2017-01-10 2018-07-19 中国科学院深圳先进技术研究院 Analysis method for multi-mode radiomics, apparatus and terminal
CN108550150A (en) * 2018-04-17 2018-09-18 上海联影医疗科技有限公司 Acquisition methods, equipment and the readable storage medium storing program for executing of breast density
CN108629761A (en) * 2018-03-12 2018-10-09 中山大学 A kind of breast cancer image-recognizing method, device and user terminal
CN108681731A (en) * 2018-04-03 2018-10-19 中山大学 A kind of thyroid cancer ultrasound picture automatic marking method and system
CN108960313A (en) * 2018-06-26 2018-12-07 南京工程学院 Ultrasonic breast lump hierarchical detection method based on Shearlet feature and level binary tree SVM classifier
CN109146848A (en) * 2018-07-23 2019-01-04 东北大学 A kind of area of computer aided frame of reference and method merging multi-modal galactophore image
CN109410167A (en) * 2018-08-31 2019-03-01 深圳大学 A kind of analysis method and Related product of 3D galactophore image
CN109598709A (en) * 2018-11-29 2019-04-09 东北大学 Mammary gland assistant diagnosis system and method based on fusion depth characteristic
CN109893100A (en) * 2019-04-18 2019-06-18 盐城工学院 A kind of method that breast density quantification calculates in breast cancer risk assessment
CN110610498A (en) * 2019-08-13 2019-12-24 上海联影智能医疗科技有限公司 Mammary gland molybdenum target image processing method, system, storage medium and equipment
CN112639997A (en) * 2018-08-27 2021-04-09 第一百欧有限公司 Disease diagnosis system and method using neural network for segmentation

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
JUNCHANG XIN等: "NMR image segmentation based on Unsupervised Extreme Learning Machine", 《PROCEEDINGS OF ELM-2015》 *
刘庆庆: "钼靶图像计算机辅助诊断方法研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 *
王梦珍: "乳腺医学图像处理方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
马萌: "基于Directionlet和稀疏表示的乳腺X线图像增强和检测", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106780451A (en) * 2016-12-07 2017-05-31 西安电子科技大学 X-ray, ultrasound, infrared image fusion breast lump image detecting method
WO2018129650A1 (en) * 2017-01-10 2018-07-19 中国科学院深圳先进技术研究院 Analysis method for multi-mode radiomics, apparatus and terminal
CN107358267A (en) * 2017-07-20 2017-11-17 东北大学 A kind of breast ultrasound image multivariate classification system and method based on cross-correlation feature
CN107392928A (en) * 2017-07-20 2017-11-24 东北大学 Breast cancer diagnosis system and method based on MRF Snake segmentations
CN108629761B (en) * 2018-03-12 2020-11-03 中山大学 Breast cancer image identification method and device and user terminal
CN108629761A (en) * 2018-03-12 2018-10-09 中山大学 A kind of breast cancer image-recognizing method, device and user terminal
CN108681731A (en) * 2018-04-03 2018-10-19 中山大学 A kind of thyroid cancer ultrasound picture automatic marking method and system
CN108550150A (en) * 2018-04-17 2018-09-18 上海联影医疗科技有限公司 Acquisition methods, equipment and the readable storage medium storing program for executing of breast density
CN108550150B (en) * 2018-04-17 2020-11-13 上海联影医疗科技有限公司 Method and device for acquiring mammary gland density and readable storage medium
CN108960313A (en) * 2018-06-26 2018-12-07 南京工程学院 Ultrasonic breast lump hierarchical detection method based on Shearlet feature and level binary tree SVM classifier
CN108960313B (en) * 2018-06-26 2021-07-02 南京工程学院 Ultrasonic breast mass grading detection method based on Shearlet feature and hierarchical binary tree SVM classifier
CN109146848A (en) * 2018-07-23 2019-01-04 东北大学 A kind of area of computer aided frame of reference and method merging multi-modal galactophore image
CN112639997A (en) * 2018-08-27 2021-04-09 第一百欧有限公司 Disease diagnosis system and method using neural network for segmentation
CN109410167A (en) * 2018-08-31 2019-03-01 深圳大学 A kind of analysis method and Related product of 3D galactophore image
CN109410167B (en) * 2018-08-31 2021-11-09 深圳大学 Analysis method, system and medium for 3D mammary gland image
CN109598709A (en) * 2018-11-29 2019-04-09 东北大学 Mammary gland assistant diagnosis system and method based on fusion depth characteristic
CN109893100A (en) * 2019-04-18 2019-06-18 盐城工学院 A kind of method that breast density quantification calculates in breast cancer risk assessment
CN109893100B (en) * 2019-04-18 2022-05-10 盐城工学院 Quantitative calculation method for breast density
CN110610498A (en) * 2019-08-13 2019-12-24 上海联影智能医疗科技有限公司 Mammary gland molybdenum target image processing method, system, storage medium and equipment

Similar Documents

Publication Publication Date Title
CN106023239A (en) Breast lump segmentation system and method based on mammary gland subarea density clustering
CN109598709B (en) Mammary gland auxiliary diagnosis system and method based on fusion depth characteristic
Al-Antari et al. Evaluation of deep learning detection and classification towards computer-aided diagnosis of breast lesions in digital X-ray mammograms
Younis et al. Classification of skin cancer dermoscopy images using transfer learning
CN104268873B (en) Breast tumor partition method based on nuclear magnetic resonance images
JP2022518446A (en) Medical image detection methods and devices based on deep learning, electronic devices and computer programs
Paul et al. Predicting nodule malignancy using a CNN ensemble approach
Kaucha et al. Early detection of lung cancer using SVM classifier in biomedical image processing
CN108122236A (en) Iterative eye fundus image blood vessel segmentation method based on distance modulated loss
CN103425986B (en) Mammary gland tumor image characteristic extracting method based on edge neighborhood weighting
Palma et al. Detection of masses and architectural distortions in digital breast tomosynthesis images using fuzzy and a contrario approaches
CN103871066B (en) The building method of similarity matrix in ultrasonoscopy Ncut segmentation
CN107358267A (en) A kind of breast ultrasound image multivariate classification system and method based on cross-correlation feature
CN101980287A (en) Method for detecting image edge by nonsubsampled contourlet transform (NSCT)
Assari et al. A bimodal BI-RADS-guided GoogLeNet-based CAD system for solid breast masses discrimination using transfer learning
Wei et al. A novel method for segmentation of CT head images
CN107392204A (en) A kind of galactophore image microcalcifications automatic checkout system and method
Li et al. Improved EfficientNet-B4 for melanoma detection
Rizzi et al. A supervised method for microcalcification cluster diagnosis
Al-Absi et al. Computer aided diagnosis system based on machine learning techniques for lung cancer
Mustafa et al. Mammography image segmentation: Chan-Vese active contour and localised active contour approach
Razali et al. High-level features in deeper deep learning layers for breast cancer classification
CN107463964A (en) A kind of tumor of breast sorting technique based on features of ultrasound pattern correlation, device
Chen et al. Enhancement of Breast Mammography to Rapid Screen Abnormalities Using 2D Spatial Fractional‐Order Feature Extraction and Multilayer Machine Vision Classifier
Carcagnì et al. A systematic investigation on deep architectures for automatic skin lesions classification

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20161012