CN107103325A - A kind of histopathology image classification method - Google Patents
A kind of histopathology image classification method Download PDFInfo
- Publication number
- CN107103325A CN107103325A CN201710259931.7A CN201710259931A CN107103325A CN 107103325 A CN107103325 A CN 107103325A CN 201710259931 A CN201710259931 A CN 201710259931A CN 107103325 A CN107103325 A CN 107103325A
- Authority
- CN
- China
- Prior art keywords
- image
- dictionary
- histopathology
- rgb triple
- dsift
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of histopathology image classification method, gray scale and RGB feature are combined, while considering the general character and difference between RGB triple channel features.First, the DSIFT features of histopathology gradation of image passage and RGB triple channels are extracted respectively, the sub- dictionary of each passage is obtained accordingly, realize the joint sparse coding of shared component and exclusive component.Then, with reference to the spatial information of image, using SPM models, multichannel combined sparse coding is carried out to the characteristics of image of different levels.Finally classified using code coefficient.Model proposed by the present invention has more preferable character representation ability, and code coefficient has more preferable classification performance, can effectively aid in the clinical diagnosis of doctor.
Description
Technical field
The present invention relates to image processing field, particularly a kind of histopathology image classification method.
Background technology
Histopathology image has become generally acknowledged " goldstandard " of WHO in medical diagnosis on disease[1], for auxiliary
Medical diagnosis on disease plays extremely important effect.However, analysis of the virologist to histopathology image rests on be based on always
The qualitative level of experience.Therefore, how computer-aided diagnosis (Computer aided diagnosis, CAD) technology is passed through
Automatically realize histopathology image quantitative analysis, it has also become one of medical image analysis area research focus[2-4]。
Traditional area of computer aided disease diagnosing system is typically entered using the gray scale or textural characteristics of histopathology image
Row medical diagnosis on disease[5-6], certain effect is achieved, and with preferable real-time.But, often ignore histopathology image
Colouring information.Srinivas etc.[7]Based on rarefaction representation sorting technique[8](Sparse representation based
Classification, SRC) a kind of synchronous sparse model is proposed, the model utilizes the pixel of histopathology image RGB triple channels
Value realizes histopathology image classification as dictionary by the reconstructed error of rarefaction representation.Vu etc.[9]Using colouring information, it will scheme
As the RGB triple channel values of block are connected into column vector as sample, it is proposed that a kind of new dictionary learning model, the model can be certainly
It is dynamic to extract identification feature, realize the classification of ill and disease-free sample.Although above-mentioned histopathology image classification method is obtained
Certain effect, does not account for the general character and otherness between histopathology image RGB triple channels but.
In recent years, Duarte etc.[10]Joint sparse model (Joint sparsity models, JSM) is proposed, and
Image co-registration is applied to rapidly[11], denoising[12], recover[13]And pattern-recognition[14-15]Deng field.Joint sparse model is divided into
Three classes：JSM-1, JSM-2 and JSM-3.JSM-1[10]Signal is divided into two parts：Shared component and exclusive component, combine same
Individual dictionary can obtain the sparse coding of two components.Meanwhile, Yang etc.[16]Using the spatial positional information of image, by golden word
Tower Matching Model[17](Spatial Pyramid Matching, SPM) is incorporated into the sparse coding of characteristics of image to improve classification
Performance.Therefore, Shi etc.[18]With reference to JSM-1 and SPM models, it is proposed that match mould based on the spatial pyramid that joint sparse is encoded
Type (Joint sparse coding based spatial pyramid matching, JScSPM) model.First, K- is utilized
Means is by all DSIFT features of RGB triple channels[19]Cluster together, shared component and exclusive component are used as using the feature of cluster
Dictionary, set up joint sparse encoding model, sparse coding coefficient obtained with reference to SPM models, and applied to histopathology image
Classification, achieves preferable classification performance.But there is problems with：Although 1) JScSPM models will using joint sparse model
Each passage is divided into shared component and exclusive component, but two parts are encoded using identical dictionary, do not only result in two points
Similarity is higher between the code coefficient of amount, and can not be between effective district subchannel exclusive component between otherness, coding system
Number identification is weak；2) JScSPM models ignore the half-tone information of image, and to the DSIFT features of all color channels together
Cluster obtains dictionary, have ignored the association between tri- passages of RGB, and its model can not really represent the shared of triple channel feature
Component and exclusive component.
The JSM that Duarte etc. [10] is proposed is broadly divided into three types：JSM-1, JSM-2 and JSM-3.Wherein, JSM-1
Signal is divided into two parts：Shared component and exclusive component, and two parts carry out sparse coding, target using identical dictionary
Function is defined as follows：
Wherein, ycIt is signal yjShared component,It is signal yjExclusive component, xcWithRespectively ycWithIn dictionary
Sparse coding coefficient under D, K and KjTo have the degree of rarefication of component and exclusive component.
Shi etc. [17] proposes JScSPM models, by two part sums of DSIFT character representations of image RGB triple channels：
Shared component and exclusive component, model definition are as follows：
Wherein, yR,yG,yBThe DSIFT features of RGB triple channels, y are corresponded to respectivelycIt is the shared component of triple channel, It is that RGB triple channels distinguish corresponding exclusive component.
Secondly, using K-means algorithms, all DSIFT features of RGB triple channels is clustered together and obtain dictionary D, used
Identical dictionary D carries out sparse coding to shared component and exclusive component, sets up joint sparse encoding model, it is defined as follows：
Wherein, xcThe code coefficient of component is had for triple channel,For the volume of tri- exclusive components of passage of RGB
Code coefficient.
Formula (3) is changed into following matrix form：
Wherein, yR,yG,yBRespectively DSIFT features of histopathology image RGB triple channels, by yR,yG,yBThree arrange to
Amount series connection is used as signal y=[yR,yG,yB]T,To combine dictionary,For joint sparse code coefficient.
Finally, it is considered to histopathology image space structural information, three layers are divided to characteristics of image, with reference to SPM models, JScSPM
Model is defined as follows：
Wherein, Y=[y1,y2,…,yN]∈RM×NFor every tomographic image DSIFT set of eigenvectors, combine dictionaryFor excessively complete dictionary,For joint sparse code coefficient.Formula (5)
In Section 1 be sparse reconstructed error, Section 2 be based on l1The sparse item of coding of norm.
JScSPM models couplings joint sparse and spatial pyramid model, realize the classification of histopathology image, obtain
Certain experiment effect.But, in JScSPM models, the half-tone information of histopathology image is ignored, also without real
Consideration RGB triple channel features between general character and difference, the important joint sparse of institute is encoded using identical dictionary, coding
The discriminating power of coefficient is weak, causes classification performance to have much room for improvement.
The content of the invention
The technical problems to be solved by the invention are, in view of the shortcomings of the prior art, providing a kind of histopathology image classification
Method.
In order to solve the above technical problems, the technical solution adopted in the present invention is：A kind of histopathology image classification method,
Comprise the following steps：
1) training image of histopathology image is inputted, the gray channel and histopathology image of training image is obtained
RGB triple channel images；
2) the DSIFT characteristic vectors of gray channel image and RGB triple channel images are extracted respectively, by RGB triple channel images
DSIFT characteristic vectors series connection, obtain training sample yi, and by all training sample yiObtain sample matrix Y；
3) the DSIFT characteristic vectors of cluster gray channel image and the DSIFT characteristic vectors of RGB triple channel images, are obtained
Corresponding sub- dictionary DI, DR、DG、DB, tectonic syntaxis dictionary；
4) RGB triple channel image feature vectors are expressed as shared component and exclusive component sum, set up multichannel combined
Sparse model；
5) training image or test image of histopathology are divided into 3 layers, this 3 layers are divided into 1H, 4H, 16H accordingly
Image block, sets up every layer of MC-JScSPM models, obtains each tomographic image sample matrix Y joint sparse code coefficient；
6) pondization operation is carried out to joint sparse coding coefficient, obtains the code coefficient of feature；
7) code coefficient is trained into grader as SVM input；
8) be based on step 5) and step 6) obtain test set joint sparse code coefficient, utilize the grader progress
Classification experiments.
The expression formula that RGB triple channel image feature vectors are expressed as into shared component and exclusive component sum is：
Wherein, yR,yG,yBThe DSIFT characteristic vectors of RGB triple channel images are corresponded to respectively；For RGBRGB
The code coefficient of the exclusive component of triple channel image feature vector；DIFor the sub- dictionary of gray scale；DR、DG、DBRespectively RGB triple channels are divided
Not corresponding sub- dictionary；D is based on for shared componentICode coefficient,ForBased on sub- dictionary
DR、DG、DBCode coefficient,It is that RGB triple channels distinguish corresponding exclusive component.
The expression formula of the multichannel combined sparse model is：
Wherein,For combined coding coefficient；Y=[yR,yG,yB]TFor histopathology image RGB threeways
The column vector that the DSIFT features in road are in series.
The MC-JScSPM model expressions are as follows：
Wherein, Y=[y1,y2,…,yN]∈RM×NIt is characterized vector set；For joint dictionary；For sparse coding coefficient；, N is input sample number, and K is the number of column vector in dictionary.
MC-JScSPM models are solved using LARS methods, sparse coding coefficient is obtained
Compared with prior art, the advantageous effect of present invention is that：The gray scale of conjunctive tissue pathological image of the present invention
Passage and color channel feature, based on this similar low priori between the exclusive component of RGB channel feature, it is proposed that one kind is based on
Spatial pyramid Matching Model (the Multi-channel joint sparse coding of multichannel combined sparse coding
Based spatial pyramid matching, MC-JScSPM) histopathology image classification method.This method is carried first
The gray channel of histopathology image and the DSIFT features of RGB triple channels are taken, the DSIFT features of each passage are clustered respectively, are obtained
It is respectively to four sub- dictionaries：Gray scale dictionary, R passages dictionary, G passages dictionary and channel B dictionary.Using gray scale dictionary as each
The shared dictionary of channel characteristics, other three sub- dictionaries respectively as each channel characteristics exclusive dictionary, and set up joint it is dilute
Dredge encoding model.Secondly, with reference to SPM models, multichannel combined sparse coding is carried out to the characteristics of image of different levels, volume is utilized
Code coefficient training SVM classifier.The present invention can effectively improve the classification performance of histopathology image, and code coefficient identification is strong；
The association between tri- passages of RGB is considered, model can really represent the shared component and exclusive component of triple channel feature.
Brief description of the drawings
Fig. 1 is MC-JScSPM model flow figures；
Fig. 2 (a) is lung, spleen, the healthy image of kidney；Fig. 2 (b) is lung, spleen, the inflammation image of kidney；
Fig. 3 (a) is the ROC characteristic curves of lung；Fig. 3 (b) is the ROC characteristic curves of spleen；Fig. 3 (c) is the ROC of kidney
Characteristic curve.
Embodiment
The present invention extracts the DSIFT features of histopathology gradation of image passage and colour RGB triple channels respectively, by cluster
Gray feature is as shared dictionary, and the RGB triple channel DSIFT features clustered respectively propose a kind of MC- as exclusive dictionary
JScSPM models.
Assuming that matrix A, B ∈ Rm×n, matrix A, B similarities are defined as follows：
Wherein,<A,B>=tr (BTA), tr (*) representing matrix leading diagonal sum；
θ represents the angle between two matrixes, and f (θ) codomain is [- 1,1].When θ=90 °, f (θ)
When=0, A is represented, B matrixes are completely dissimilar；When θ=0 °, f (θ)=1, A, B matrix similarity highests are represented.
The present invention is using gray channel feature first as shared component, and convolution (2) tries to achieve the exclusive of RGB triple channel features
Component, the similarity between exclusive component is calculated using formula (6).In order to verify between tissue pathological image RGB triple channels exclusive point
The similitude of amount, based on ADL data sets[20]In three class loading pathological images, respectively calculate RGB triple channels DSIFT features it is only
Similarity between important, as a result as shown in table 1.
Table 1：Similarity between the exclusive component of RGB triple channel DSIFT features
As known from Table 1, the similarity between the exclusive component of RGB triple channel DSIFT features of this three class loadings pathological image
Very low, illustrating the sparse coding coefficient of the exclusive component of RGB triple channel features has preferable identification, can obtain and preferably divide
Class result.
The present invention proposes MC-JScSPM models, with document[18]Unlike, the present invention is using K-means respectively to ash
Degree passage, RGB triple channel features are clustered, and obtain the corresponding sub- dictionary D of each passageI, DR、DG、DB.Wherein, the sub- word of gray scale
Allusion quotation DIDictionary is had as RGB triple channels feature, by DR、DG、DBIt is used as the exclusive dictionary of RGB triple channel features.Utilize sub- dictionary DI,
DR、DG、DBTectonic syntaxis dictionary, sets up the sparse model of multichannel, and formula (2) can redefine as follows：
Wherein,To have component；Exclusive point of respectively each channel characteristics
Amount；DIFor the sub- dictionary of gray scale, DR、DG、DBRespectively RGB triple channels distinguish corresponding sub- dictionary,D is based on for shared componentI's
Code coefficient,ForBased on sub- dictionary DR、DG、DBCode coefficient.
Then, formula (7) is converted into matrix form, it is defined as follows：
Wherein,To combine dictionary,For combined coding coefficient, y
=[yR,yG,yB]TThe column vector being in series for the DSIFT features of histopathology image RGB triple channels.
Finally, with reference to image space structural information, using SPM models, MC-JScSPM models is set up, are defined as follows：
Wherein, Y=[y1,y2,…,yN]∈RM×NIt is characterized vector set,To combine dictionary,For sparse coding coefficient.The present invention utilizes LARS[21]Solution formula (9) can obtain joint sparse coding system
Number
The present invention proposes the flow of model as shown in figure 1, specific steps are described as follows：
Step 1：Training sample is inputted, the gray channel and RGB triple channel images of training sample is respectively obtained.
Step 2：Gray channel and the DSIFT characteristic vectors of RGB triple channels are extracted respectively, by RGB triple channel DSIFT features
Vector series connection obtains training characteristics yi, combine and obtain eigenmatrix Y.
Step 3：Cluster gray scale, the DSIFT features of RGB triple channels, obtain corresponding sub- dictionary D respectivelyI, DR、DG、DB, and
For tectonic syntaxis dictionary.
Step 4：According to formula (7), it is shared component and exclusive component sum by RGB triple channels character representation, utilizes step 3
The joint dictionary of construction, multichannel combined sparse model is set up according to formula (8).
Step 5：With reference to spatial pyramid structure, image is divided into 3 layers, and be divided into 1 accordingly, 4,16 image blocks,
Every layer of MC-JScSPM models are set up according to formula (9), LARS algorithms are utilized[21]Solution formula (9) obtains each tomographic image feature
Joint sparse code coefficient.
Step 6：Pondization operation is carried out to joint sparse coding coefficient using multiple dimensioned maximum pond method, feature is obtained
Code coefficient.
Step 7：The feature coding coefficient obtained using step 6, as SVM input, trains grader.
Step 8：The code coefficient of test sample is obtained using step 5,6, and utilizes the SVM trained using step 7 kind
Grader carries out classification experiments.
The present invention is based on ADL[20]Histopathology image data set, demonstrates the validity of MC-JSCSPM models, and and its
He is analyzed method.
DSIFT features[19]Extraction process be：Using grid, image is divided to the characteristic block for obtaining formed objects, and
Overlap mode is used between block and block, the center of each characteristic block is as a characteristic point, by same characteristic block
All pixels point is as the SIFT feature of this feature point, and it is then DSIFT features that the SIFT of all characteristic points of image, which is combined,.
The sampling interval of DSIFT features of the present invention is set to 16 pixels, and tile size is 16 × 16, then piece image
DSIFT eigenmatrixes size be 128*1369.
ADL data sets and related experiment are set：
ADL data sets[20]There is provided by Pennsylvania State University, have 900 multiple images, include the figure of three organoids
Picture：Lung, spleen, kidney.Wherein, each organoid all includes health and the other pathological tissue of the species of inflammation two, respectively accounts for more than 150
, its size is 1360 × 1024.Two kinds of medical pathologies schematic illustration of tissue of each organoid as shown in Fig. 2 Fig. 2 (a) from a left side to
Right to represent lung, spleen, the healthy image of kidney respectively successively, Fig. 2 (b) represents lung, spleen, the inflammation of kidney successively from left to right
Image.
The present invention is randomly selected respectively from the color catalog image of lung, spleen, the health of the organoid of kidney three and inflammation
120, as training set, by every coloured image Size Conversion into 600 × 600, extract its gray channel and RGB tri- respectively
The DSIFT features of passage, then the DSIFT of each passage be characterized as 128 × 1369 × 120, randomly select 10000 from each passage
Individual DSIFT features are clustered, and obtain sub- dictionary DI, DR、DG、DB∈R128×1024, utilize sub- dictionary tectonic syntaxis dictionaryDivide three layers by every image, column vector is connected into per tomographic image RGB triple channel features as training sample, profit
With joint dictionarySparse coding is carried out, code coefficient is subjected to maximum pond, Chi Huahou SVM points of code coefficient training is utilized
Class device.Finally, every coloured image size in test set is equally changed into 600 × 600, is divided into three layers, extracted per tomographic image
The DSIFT features of RGB triple channels, are connected into column vector as test sample, using combining dictionaryObtain the dilute of test sample
Code coefficient is dredged, same maximum pondization operation is carried out, and classified using SVM classifier.
The experimental result comparative analysis of the inventive method and other method is as follows：
In order to prove that the present invention puies forward the validity of MC-JScSPM methods, we compared for ScSPM and JScSPM methods.
Table 2,3,4 sets forth classification results of the distinct methods on lung, spleen and renal image.
The distinct methods of table 2 are contrasted in the classification results of lung images
The distinct methods of table 3 are contrasted in the classification results of spleen image
The distinct methods of table 4 are contrasted in the classification results of renal image
From table 2, table 3 and table 4, MC-JScSPM models proposed by the present invention are to lung, spleen and the organoid of kidney three
Medical diagnosis on disease effect be substantially better than other models.The present invention combines the gray scale and colour information of image, using gray scale dictionary
With tri- dictionaries of RGB as combining dictionary, it is considered to the association between multichannel, take full advantage of general character between different passages with
Difference, the identification of obtained combined coding coefficient is stronger, obtains more preferable classifying quality.
The ROC characteristic curves of the inventive method and other method are analyzed as follows：
Fig. 3 (a), Fig. 3 (b), Fig. 3 (c) divide the ROC curve comparing result for giving the inventive method and other method, Fig. 3
(a), Fig. 3 (b) and Fig. 3 (c) is respectively the ROC performance diagrams of lung, spleen and kidney, it can be seen that of the invention
The ROC curve of method further illustrates the superiority that the present invention puies forward MC-JScSPM methods apparently higher than other method.
The present invention proposes a kind of spatial pyramid Matching Model based on multichannel combined sparse coding, and is applied
In the classification of histopathology image.The model not only combines gray scale and RGB feature, at the same consider RGB triple channels feature it
Between general character and difference.First, the DSIFT features of histopathology gradation of image passage and RGB triple channels are extracted respectively, accordingly
The sub- dictionary of each passage is obtained, the joint sparse coding of shared component and exclusive component is realized.Then, with reference to the sky of image
Between information, using SPM models, multichannel combined sparse coding is carried out to the characteristics of image of different levels.Finally utilize coding system
Number is classified.Test result indicates that, model proposed by the present invention has more preferable character representation ability, and code coefficient has more
Good classification performance, can effectively aid in the clinical diagnosis of doctor.
Bibliography：
[1]CHAN J K.The wonderful colors of the hematoxylin-eosin stain in
diagnostic surgical pathology.[J].International Journal of Surgical
Pathology,2014,22(1):12-32.
[2]VETA M,PLUIM J P,VAN DIEST P J,et al.Breast cancer histopathology
image analysis:a review.[J].IEEE transactions on bio-medical engineering,
2014,61(61):1400-11.
[3]RITTER S,MARGULIE K B.Emerging tools for computer-aided diagnosis
and prognostication[J].Journal of Clinical Trials,2011,4(2):e117-e117.
[4]XIAN M,ZHANG Y,CHENG H D.Fully automatic segmentation of breast
ultrasound images based on breast characteristics in space and frequency
domains[J].Pattern Recognition,2015,48(2):485-497
[5]XU J,JANOWCZYK A.A high-throughput active contour scheme for
segmentation of histopathological imagery.[J].Medical Image Analysis,2011,15
(6):851-862.
[6]KONG J,COOPER LA,WANG F,GUTMAN DA,GAO J,CHISOLM C,et
al.Integrative mul-timodal analysis of glioblastoma using TCGA molecular data
pathology images and clinical outcomes.IEEE Trans Biomed Eng 2011；58(12):
3469–74.
[7]SRINIVAS U,MOUSAVI H S,MONGA V,et al.Simultaneous sparsity model
for histopathological image representation and classification[J].IEEE
transactions on medical imaging,2014,33(5):1163-1179.
[8]WRIGHT J,YANG A Y,GANESH A,et al.Robust face recognition via
sparse representation[J].IEEE Transactions on Pattern Analysis&Machine
Intelligence,2009,31(2):210-227.
[9]VU T H,MOUSAVI H S,MONGA V,et al.DFDL:Discriminative feature-
oriented dictionary learning for histopathological image classification[C],
2015 IEEE 12th International Symposium on Biomedical Imaging(ISBI).IEEE,2015:
990-994.
[10]DUARTE M F,WAKIN MB,BARON D,BARANIUK RG.Universal distributed
sensing via random projections.In:Proceedings of the symposium on information
pro-cessing in sensor networks.2006.p.177–85.
[11]YU N,QIU TS,BI F,WANG AQ.Image features extraction and fusion
based on joint sparse representation.IEEE J Sel Top Signal Process 2011；5(5):
1074–82
[12]YU N,QIU T,REN FQ.Denoising for multiple image copies through
joint sparse representation.J Math Imaging Vision 2013；45(1):46–54.
[13]LUO J,YANG B,CHEN Z.Color image restoration via extended joint
sparse model.Commun Comput Inf Sci 2012；321:497–504.
[14]RAMEZANI M,ABOLMAESUMI P,MARBLE K,et al.Joint sparse
representation of brain activity patterns related to perceptual and cognitive
components of a speech comprehension task[C]//International Workshop on
Pattern Recognition in Neuroimaging.IEEE,2012:29-32.
[15]SHEKHAR S,PATEL V M,NASRABADI N M,et al.Joint sparsity-based
robust multimodal biometrics recognition[C]//International Conference on
Computer Vision.Springer-Verlag,2012:365-374.
[16]YANG J,YU K,GONG Y,et al.Linear spatial pyramid matching using
sparse coding for image classification[J].2009:1794-1801.
[17]LAZEBNIK S,SCHMID C,PONCE J.Beyond bags of features:Spatial
pyramid matching for recognizing natural scene categories[C]//2006 IEEE
Computer Society Conference on Computer Vision and Pattern Recognition(CVPR'
06).IEEE,2006,2:
[18]SHI J,LI Y,ZHU J,et al.Joint sparse coding based spatial pyramid
matching for classification of color medical image[J].Computerized Medical
Imaging&Graphics the Official Journal of the Computerized Medical Imaging
Society,2015,41(1):61-66.
[19]LI F F,PERONA P.A bayesian hierarchical model for learning
natural scene categories[C]//IEEE Computer Society Conference on Computer
Vision and Pattern Recognition.IEEE Computer Society,2005:524-531 vol.2.
[20]MONGA V,Animal diagnostics lab dataset,http://vbs.psu.edu/adl,
2013
[21]EFRON B.Least angle regression[J].Mathematics,2004,32(2):2004.
Claims (6)
1. a kind of histopathology image classification method, it is characterised in that comprise the following steps：
1) training image of histopathology image is inputted, the gray channel of training image and the RGB tri- of histopathology image is obtained
Channel image；
2) the DSIFT characteristic vectors of gray channel image and RGB triple channel images are extracted respectively, by RGB triple channel images
DSIFT characteristic vectors are connected, and obtain training sample yi, and by all training sample yiObtain sample matrix Y；
3) the DSIFT characteristic vectors of cluster gray channel image and the DSIFT characteristic vectors of RGB triple channel images, obtain corresponding
Sub- dictionary DI, DR、DG、DB, tectonic syntaxis dictionary；
4) RGB triple channel image feature vectors are expressed as shared component and exclusive component sum, set up multichannel combined sparse
Model；
5) training image or test image of histopathology are divided into 3 layers, this 3 layers are divided into 1H, 4H, 16H images accordingly
Block, sets up every layer of MC-JScSPM models, obtains each tomographic image sample matrix Y joint sparse code coefficient；
6) pondization operation is carried out to joint sparse coding coefficient, obtains the code coefficient of feature；
7) code coefficient is trained into grader as SVM input；
8) be based on step 5) and step 6) acquisition test set joint sparse code coefficient, classified using the grader
Experiment.
2. histopathology image classification method according to claim 1, it is characterised in that by RGB triple channel characteristics of image
Vector representation is that the expression formula of shared component and exclusive component sum is：
Wherein, yR,yG,yBThe DSIFT characteristic vectors of RGB triple channel images are corresponded to respectively；For RGB triple channel images
The code coefficient of the exclusive component of characteristic vector；DIFor the sub- dictionary of gray scale；DR、DG、DBRespectively RGB triple channels distinguish corresponding son
Dictionary；D is based on for shared componentICode coefficient,For Based on sub- dictionary DR、DG、DBVolume
Code coefficient,It is that RGB triple channels distinguish corresponding exclusive component.
3. histopathology image classification method according to claim 2, it is characterised in that the multichannel combined sparse mould
The expression formula of type is：
Wherein,For combined coding coefficient；Y=[yR,yG,yB]TFor histopathology image RGB triple channels
The column vector that DSIFT features are in series.
4. histopathology image classification method according to claim 1, it is characterised in that H=1.
5. histopathology image classification method according to claim 3, it is characterised in that the MC-JScSPM model tables
It is as follows up to formula：
Wherein, Y=[y1,y2,…,yN]∈RM×NIt is characterized vector set；For joint dictionary；For sparse coding coefficient, λ is regularization coefficient, and N is input sample number, and K is column vector in dictionary
Number.
6. histopathology image classification method according to claim 4, it is characterised in that solve MC- using LARS methods
JScSPM models, obtain sparse coding coefficient
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710259931.7A CN107103325B (en) | 2017-04-20 | 2017-04-20 | Tissue pathology image classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710259931.7A CN107103325B (en) | 2017-04-20 | 2017-04-20 | Tissue pathology image classification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107103325A true CN107103325A (en) | 2017-08-29 |
| CN107103325B CN107103325B (en) | 2021-01-26 |
Family
ID=59657139
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710259931.7A Active CN107103325B (en) | 2017-04-20 | 2017-04-20 | Tissue pathology image classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107103325B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107595390A (en) * | 2017-10-19 | 2018-01-19 | 青岛大学附属医院 | A kind of real-time matching fusion method of ultrasonic image and CT images |
| CN109063766A (en) * | 2018-07-31 | 2018-12-21 | 湘潭大学 | A kind of image classification method based on identification prediction sparse decomposition model |
| CN114648531A (en) * | 2022-05-20 | 2022-06-21 | 领伟创新智能系统(浙江)有限公司 | Solar panel surface dust identification method based on color channel brightness distribution |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103336835A (en) * | 2013-07-12 | 2013-10-02 | 西安电子科技大学 | Image retrieval method based on weight color-sift characteristic dictionary |
| CN104766096A (en) * | 2015-04-17 | 2015-07-08 | 南京大学 | Image classification method based on multi-scale global features and local features |
| CN105740772A (en) * | 2016-01-22 | 2016-07-06 | 江苏师范大学 | Real time monitoring method of cortical somatosensory evoked potential of transform joint sparse model |
| CN106339686A (en) * | 2016-08-29 | 2017-01-18 | 江苏大学 | Pedestrian detection method based on sparse discrimination template (SDT) features |
-
2017
- 2017-04-20 CN CN201710259931.7A patent/CN107103325B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103336835A (en) * | 2013-07-12 | 2013-10-02 | 西安电子科技大学 | Image retrieval method based on weight color-sift characteristic dictionary |
| CN104766096A (en) * | 2015-04-17 | 2015-07-08 | 南京大学 | Image classification method based on multi-scale global features and local features |
| CN105740772A (en) * | 2016-01-22 | 2016-07-06 | 江苏师范大学 | Real time monitoring method of cortical somatosensory evoked potential of transform joint sparse model |
| CN106339686A (en) * | 2016-08-29 | 2017-01-18 | 江苏大学 | Pedestrian detection method based on sparse discrimination template (SDT) features |
Non-Patent Citations (2)
| Title |
|---|
| JUN SHI ET AL.: "Joint sparse coding based spatial pyramid matching for classification of color medical image", 《COMPUTERIZED MEDICAL IMAGING & GRAPHICS》 * |
| 祝璞: "基于多通道的分层特征提取的图像识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107595390A (en) * | 2017-10-19 | 2018-01-19 | 青岛大学附属医院 | A kind of real-time matching fusion method of ultrasonic image and CT images |
| CN107595390B (en) * | 2017-10-19 | 2020-12-08 | 青岛大学附属医院 | Real-time matching fusion method of ultrasonic image and CT image |
| CN109063766A (en) * | 2018-07-31 | 2018-12-21 | 湘潭大学 | A kind of image classification method based on identification prediction sparse decomposition model |
| CN114648531A (en) * | 2022-05-20 | 2022-06-21 | 领伟创新智能系统(浙江)有限公司 | Solar panel surface dust identification method based on color channel brightness distribution |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107103325B (en) | 2021-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jadoon et al. | Three-class mammogram classification based on descriptive CNN features | |
| Nandi et al. | Principal component analysis in medical image processing: a study | |
| Rao et al. | Brain tumor segmentation with deep learning | |
| Tang et al. | An augmentation strategy for medical image processing based on statistical shape model and 3D thin plate spline for deep learning | |
| Gao et al. | A deep learning based approach to classification of CT brain images | |
| CN109584244B (en) | Hippocampus segmentation method based on sequence learning | |
| Li et al. | Research on the auxiliary classification and diagnosis of lung cancer subtypes based on histopathological images | |
| CN110826389B (en) | Gait recognition method based on attention 3D frequency convolution neural network | |
| CN112150428A (en) | Medical image segmentation method based on deep learning | |
| CN105760874A (en) | CT image processing system and method for pneumoconiosis | |
| CN107103325A (en) | A kind of histopathology image classification method | |
| CN111260639A (en) | Multi-view information-collaborative breast benign and malignant tumor classification method | |
| CN116030325A (en) | Lung nodule CT image recognition method based on deep hybrid learning framework | |
| CN111798424B (en) | Medical image-based nodule detection method and device and electronic equipment | |
| Qayyum et al. | Automatic segmentation using a hybrid dense network integrated with an 3D-atrous spatial pyramid pooling module for computed tomography (CT) imaging | |
| Celik et al. | A new 3D MRI segmentation method based on Generative Adversarial Network and Atrous Convolution | |
| El-Shafai et al. | An efficient medical image deep fusion model based on convolutional neural networks | |
| Gopikha et al. | Regularised Layerwise Weight Norm Based Skin Lesion Features Extraction and Classification. | |
| Zhao et al. | Segmenting brain tissues from Chinese visible human dataset by deep-learned features with stacked autoencoder | |
| Gong et al. | Towards accurate pulmonary nodule detection by representing nodules as points with high-resolution network | |
| Pavithra et al. | An Overview of Convolutional Neural Network Architecture and Its Variants in Medical Diagnostics of Cancer and Covid-19 | |
| Malathi et al. | Active contour based segmentation and classification for pleura diseases based on Otsu’s thresholding and support vector machine (SVM) | |
| Merati et al. | A New Triplet Convolutional Neural Network for Classification of Lesions on Mammograms. | |
| CN115222637A (en) | Multi-modal medical image fusion method based on global optimization model | |
| Hussain et al. | RecU-Net++: Improved utilization of receptive fields in U-Net++ for skin lesion segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |