CN109410196A - Cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field - Google Patents
Cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field Download PDFInfo
- Publication number
- CN109410196A CN109410196A CN201811242279.9A CN201811242279A CN109410196A CN 109410196 A CN109410196 A CN 109410196A CN 201811242279 A CN201811242279 A CN 201811242279A CN 109410196 A CN109410196 A CN 109410196A
- Authority
- CN
- China
- Prior art keywords
- image
- cervical cancer
- cancer tissues
- digitlization
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20152—Watershed segmentation
-
- 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
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The present invention relates to a kind of cervical cancer tissues pathological image diagnostic methods based on Poisson annular condition random field, which comprises 101, acquisition digitlization cervical cancer tissues pathological image；102, the digitlization cervical cancer tissues pathological image of acquisition is pre-processed；103, cluster segmentation and piecemeal are carried out to by pretreated digitlization cervical cancer tissues pathological image using partitioning algorithm, obtains multiple images fritter；104, feature is extracted in each image fritter obtained from step 103, feature selecting then is carried out to the feature extracted；105, digitlization cervical cancer tissues pathological image classification is carried out to the feature chosen using conditional random field models, obtains the classification results of cervical cancer tissues pathological image.Cervical cancer tissues pathological image diagnostic method provided by the invention based on Poisson annular condition random field can obtain the classification results of cervical carcinoma according to cervical cancer tissues pathological image.
Description
Technical field
The invention belongs to cervical cancer tissues pathological image diagnostic techniques fields, and in particular to one kind is based on Poisson annular condition
The cervical cancer tissues pathological image diagnostic method of random field.
Background technique
The scheme of 1.1 prior arts is sketched
Prior art use condition random field classifies to cervicovaginal mirror image, and histopathology image is as one
Basic fact.As shown in Figure 1, this method from top to bottom includes five steps altogether:
(1) pretreatment includes image calibration, image registration and dissection feature extraction three parts；
(2) image segmentation, which uses, is based on k means clustering method, identifies it is in each organization type of homogeneity in color and intensity
Subregion；
(3) include but is not limited to Acetowhitening, inlay, dotted and atypia blood vessel diagnosis correlated characteristic；
(4) classifier of the CRF model based on the classification results according to probabilistic manner combination adjacent area: not using n
With feature f1, f2, f3 ... the characteristic function (Y1, Y2, Y3 and Y4) of fn have scaly epithelium, columnar epithelium, zone of transformation and palace
Four kinds of cluster situations (W1, W2, W3 and W4) of four kinds of histological types of neck mouth.From K arest neighbors (KNN) classifier and linearly
All results of differential analysis (LDA) classifier determine the conditional probability distribution (it is assumed that only diagnostic characteristic), after maximum
(MAP) estimation is tested to determine the parameter of Posterior probability distribution.
(5) the new mode based on window is used to determine the sensitivity and specificity of detection and diagnosis algorithm.
The objective disadvantage of 1.2 prior arts
(1) prior art, cervical cancer tissues pathological image are intended only as basic fact, it is still necessary to veteran pathology
Scholar judges image, however different pathological scholar or the same virologist of different time are to same pathological image
Judgement is also variant, and there may be biggish errors for this.
(2) prior art still needs to veteran virologist and judges, but the limited amount of virologist, and
And low developed area medical resource lacks, virologist is equally rare；Experience medico not abundant enough or virologist, to group
Reliable judgement cannot be made by knitting pathological image.
(3) prior art is only capable of judging whether regional area is abnormal using a classifier based on condition random field, and
Cannot provide the classification results of cervical carcinoma, for example, precancerous lesion (CINI, CINII, CINIII three-level) and malignant tumour it is (high, normal, basic
Break up three-level).
Summary of the invention
(1) technical problems to be solved
In order to solve the above problem of the prior art, the present invention provides a kind of uterine neck based on Poisson annular condition random field
Cancerous tissue pathological image diagnostic method can obtain the classification results of cervical carcinoma according to cervical cancer tissues pathological image.
(2) technical solution
In order to achieve the above object, the present invention uses main technical schemes the following steps are included:
A kind of cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field, which comprises
101, digitlization cervical cancer tissues pathological image is obtained；
102, the digitlization cervical cancer tissues pathological image of acquisition is pre-processed；
103, cluster segmentation is carried out to by pretreated digitlization cervical cancer tissues pathological image using partitioning algorithm
And piecemeal, obtain multiple images fritter；
104, feature is extracted in each image fritter obtained from step 103, then the feature extracted is carried out special
Sign selection；
105, digitlization cervical cancer tissues pathological image point is carried out to the feature chosen using conditional random field models
Grade obtains the classification results of cervical cancer tissues pathological image；
The conditional random field models are to be obtained using condition random field and digitlization cervical cancer tissues pathological image training
The model obtained.
Preferably, the format of the digitlization cervical cancer tissues pathological image obtained in the step 101 includes: *
.bmp、*.BMP、*.dip、*DIP、*.jpg、*.JPG、*.jpeg、*JPEG、*.jpe、*.JPE、*.jfif、*JFIF、*
.gif, * .GIF, * .GIF, * .GIF, * .GIFf, * .GIFF, * .png and * .PNG.
Preferably, the step 102 further includes following steps:
102A, image denoising is carried out to digitlization cervical cancer tissues pathological image using median filter；
102B, will be enhanced by the digitlization cervical cancer tissues pathological image of image denoising by histogram equalization
The contrast of image.
Preferably, the step 103 further includes following steps:
103A, using the image partition method clustered based on K-means to pass through pretreated digitlization cervical carcinoma group
It knits pathological image and carries out cluster segmentation；
Wherein, setting cluster numbers K is 4, image gather for nucleus, cytoplasm, cytoplasm, image labeling it is these four types of,
In conjunction with morphological operation cellulation core binary map appropriate；
103B, nucleus is positioned according to the result of image segmentation, obtains nucleus barycentric coodinates, and the gray scale of original image
Figure is divided into the image fritter of 100 × 100 pixels.
Preferably, the step 104 further includes following steps:
104A, it extracts each image fritter and is averaged DAISY feature conduct of the DAISY feature descriptor as the image fritter
Local feature；
104B, extraction full figure textural characteristics and full figure are averaged DAISY feature as global characteristics；
104C, handled using local feature and global characteristics of the Principal Component Analysis PCA to acquisition, obtain feature to
Amount group.
Preferably, the step 105 further include:
The described eigenvector group input condition random field models that selection obtains are classified, and export cervical cancer tissues
The classification results of pathological image.
Preferably, described to be indicated based on the classifier of condition random field with following formula:
Wherein, c indicates that class label, x are input picture, and Z (θ, x) is the partition functions that distribution is normalized, θ=
{θψ, θφ, θλIt is model parameter, and i corresponds to the index in figure in the image of grid, n is full graphics image fritter number.
Preferably, described eigenvector group includes: that the average DAISY feature vector of full figure nucleus, the overall situation of full figure are flat
The grey level histogram vector of equal DAISY feature vector and full grayscale image.
Preferably, it when being pre-processed in the step 102 to the digitlization cervical cancer tissues pathological image of acquisition, adopts
With mean filter method, image denoising is carried out to the digitlization cervical cancer tissues pathological image.
The partitioning algorithm used in the step 103 is watershed algorithm；
Feature is carried out to the feature extracted using principal component analytical method or Fisher face in the step 103
Selection.
Preferably, the conditional random field models include: differentiated model, middle differentiation model and low differentiation model；
The differentiated model is being averaged for the feature vector group of the digitlization cervical cancer tissues pathological image of differentiated
Value；
The middle differentiation model is being averaged for the feature vector group of the digitlization cervical cancer tissues pathological image of middle differentiation
Value；
The low differentiation model is being averaged for the feature vector group of poorly differentiated digitlization cervical cancer tissues pathological image
Value；
The conditional random field models can be according to the feature vector of the digitlization cervical cancer tissues pathological image of input
Group calculates likelihood with the differentiated model, middle differentiation model and low differentiation model respectively, and it is highest finally to choose likelihood
Model classification is as final classification results；
The classification results include: differentiated, middle differentiation and low differentiation.
(3) beneficial effect
The beneficial effects of the present invention are: the present invention provides a kind of cervical cancer tissues disease based on Poisson annular condition random field
Image diagnosing method is managed, condition random field classifier can be applied to automatically analyzing for cervical cancer tissues pathological image, be formed
The auto-check system of a set of cervical carcinoma micro-image.It is uneven, no that the present invention can suitably slow down virologist resource allocation
Enough abundant problems, can also instruct medico rationally to be diagnosed with the insufficient doctor of experience.The present invention also can solve
Contradictory problems between different doctors and under same doctor's different conditions give patient one more reliable diagnostic result.The present invention
3 ranks such as the pathological state of patient, including differentiated, middle differentiation, low differentiation more can also be accurately provided, diagnosis is improved
Accuracy rate.
Detailed description of the invention
Fig. 1 is the flow diagram of the prior art in background of invention；
Fig. 2 is a kind of cervical cancer tissues pathological image diagnosis based on Poisson annular condition random field in the embodiment of the present invention
The method flow diagram of method；
Fig. 3 is a kind of cervical cancer tissues pathological image diagnosis based on Poisson annular condition random field in the embodiment of the present invention
The method flow diagram of method；
Fig. 4 is a kind of cervical cancer tissues pathological image diagnosis based on Poisson annular condition random field in the embodiment of the present invention
Image block schematic diagram in method；
Fig. 5 is a kind of cervical cancer tissues pathological image diagnosis based on Poisson annular condition random field in the embodiment of the present invention
Simple linear chain condition random field figure in method；
Fig. 6 is a kind of cervical cancer tissues pathological image diagnosis based on Poisson annular condition random field in the embodiment of the present invention
Multilayer Poisson annular layout schematic diagram in method；
Fig. 7 is a kind of cervical cancer tissues pathological image diagnosis based on Poisson annular condition random field in the embodiment of the present invention
Analysis flow chart diagram in method.
Specific embodiment
In order to preferably explain the present invention, in order to understand, with reference to the accompanying drawing, by specific embodiment, to this hair
It is bright to be described in detail.Specific embodiments of the present invention will be described in detail with reference to the accompanying drawing.
Embodiment one
As shown in Figure 2: present embodiment discloses a kind of cervical cancer tissues pathology figures based on Poisson annular condition random field
As diagnostic method, which comprises
101, digitlization cervical cancer tissues pathological image is obtained.
102, the digitlization cervical cancer tissues pathological image of acquisition is pre-processed.
103, cluster segmentation is carried out to by pretreated digitlization cervical cancer tissues pathological image using partitioning algorithm
And piecemeal, obtain multiple images fritter.
104, feature is extracted in each image fritter obtained from step 103, then the feature extracted is carried out special
Sign selection.
105, digitlization cervical cancer tissues pathological image classification is carried out to the feature chosen using condition random field, obtained
Obtain the classification results of cervical cancer tissues pathological image.
It is noted that the digitlization cervical cancer tissues pathological image obtained in step 101 described in the present embodiment
Format includes: * .bmp, * .BMP, * .dip, * DIP, * .jpg, * .JPG, * .jpeg, * JPEG, * .jpe, * .JPE, * .jfif, *
JFIF, * .gif, * .GIF, * .GIF, * .GIF, * .GIFf, * .GIFF, * .png and * .PNG.
For example, including the poorly differentiated cervical cancer tissues pathological image data of 307 senior middle schools using one in the present embodiment
Library carries out systematic training, and the high, normal, basic differentiation of training set each 20, test set 247 is opened, and every picture size is 2560 × 1920 pictures
Element.
In detail, step 102 described in the present embodiment further includes following steps:
102A, image denoising is carried out to digitlization cervical cancer tissues pathological image using median filter.
102B, will be enhanced by the digitlization cervical cancer tissues pathological image of image denoising by histogram equalization
The contrast of image.
It should be noted that step 103 described in the present embodiment further includes following steps:
103A, using the image partition method clustered based on K-means to pass through pretreated digitlization cervical carcinoma group
It knits pathological image and carries out cluster segmentation.
Wherein, setting cluster numbers K is 4, image gather for nucleus, cytoplasm, cytoplasm, image labeling it is these four types of,
In conjunction with morphological operation cellulation core binary map appropriate.
103B, nucleus is positioned according to the result of image segmentation, obtains nucleus barycentric coodinates, and the gray scale of original image
Figure is divided into the image fritter of 100 × 100 pixels.
Step 104 described in the present embodiment further includes following steps:
104A, it extracts each image fritter and is averaged DAISY feature conduct of the DAISY feature descriptor as the image fritter
Local feature.
104B, extraction full figure textural characteristics and full figure are averaged DAISY feature as global characteristics.
104C, handled using local feature and global characteristics of the Principal Component Analysis PCA to acquisition, obtain feature to
Amount group.
It is noted that DAISY is local image characteristics description that can quickly calculate extracted towards dense characteristic, it
As essential idea with SIFT is: block statistics gradient orientation histogram, unlike, DAISY is carried out on partition strategy
It improves, carries out using Gaussian convolution the piecemeal convergence of gradient orientation histogram, utilize can quickly counting for Gaussian convolution in this way
The property calculated quickly can densely carry out the extraction of Feature Descriptor.
Step 105 described in the present embodiment further include:
The described eigenvector group input condition random field models that selection obtains are classified, and export cervical cancer tissues
The classification results of pathological image.
Specifically, it is indicated based on the classifier of condition random field with following formula described in the present embodiment:
Wherein, c indicates that class label, x are input picture, and Z (θ, x) is the partition functions that distribution is normalized, θ=
{θψ, θφ, θλIt is model parameter, and i corresponds to the index in figure in the image of grid, n is full graphics image fritter number.
It is noted that feature vector group described in the present embodiment include: full figure nucleus average DAISY feature to
Amount, the overall situation of full figure are averaged the grey level histogram vector of DAISY feature vector and full grayscale image.
The digitlization cervical cancer tissues pathological image of acquisition is pre-processed in step 102 described in the present embodiment
When, using mean filter method, image denoising is carried out to the digitlization cervical cancer tissues pathological image.
The partitioning algorithm used in the step 103 is watershed algorithm.
Feature is carried out to the feature extracted using principal component analytical method or Fisher face in the step 103
Selection.
Finally, it should be noted that conditional random field models described in the present embodiment include: differentiated model, middle differentiation model
With low differentiation model.
The differentiated model is being averaged for the feature vector group of the digitlization cervical cancer tissues pathological image of differentiated
Value.
The middle differentiation model is being averaged for the feature vector group of the digitlization cervical cancer tissues pathological image of middle differentiation
Value.
The low differentiation model is being averaged for the feature vector group of poorly differentiated digitlization cervical cancer tissues pathological image
Value.
The conditional random field models can be according to the feature vector of the digitlization cervical cancer tissues pathological image of input
Group calculates likelihood with the differentiated model, middle differentiation model and low differentiation model respectively, and it is highest finally to choose likelihood
Model classification is as final classification results.
The classification results include: differentiated, middle differentiation and low differentiation.
In detail, the classification in the present embodiment about feature extraction, selection and conditional random field models is specific as follows:
1, DAISY feature extraction is carried out to each image fritter.
2, the DAISY descriptor addition of all characteristic points in the image fritter is averaged, it is corresponding obtains the fritter
The DAISY feature vector of one same latitude.
3, the layout of application conditions random field, the DAISY feature vector of fritter is (by 0.6 where calculating each nucleus
Times be averaged DAISY feature vector and 0.25 times of middle ring of inner ring be averaged DAISY feature vector and 0.15 times of outer ring is averaged
The sum of DAISY feature vector obtains the final DAISY feature vector of blue fritter), then calculate all nucleus fritters of full figure
Averaged feature vector, the average DAISY feature vector of the available figure nucleus, i.e. A.
4, DAISY feature is extracted to full grayscale image, and is averaged, global average DAISY feature vector, i.e. B can be obtained.
5, the feature vector of one 256 dimension, i.e. C can be obtained in the grey level histogram vector for taking full grayscale image.
6, respectively to every image zooming-out A, B, C, 3 feature vectors of training set.
7, A, B, C, 3 feature vectors of differentiated digitlization cervical cancer tissues pathological image are averaged as height
The feature of differentiation model, middle differentiation and low differentiation are similarly.
8, A, B, C, 3 feature vectors of input picture are extracted.
9, A, B, C, 3 feature vectors of input picture respectively with differentiated model, middle differentiation model and low differentiation model
Likelihood is calculated, selects that class of likelihood highest as final classification result.
It is noted that training set described here refers to multiple for constructing the digitlization uterine neck of conditional random field models
Cancerous tissue pathological image, wherein the digitlization cervical cancer tissues pathological image obtained in step 101 is for testing or in fact
Digitlization cervical cancer tissues pathological image in the application of border for diagnostic imaging classification.
Embodiment two
As shown in Figure 3: the present embodiment discloses a kind of cervical cancer tissues pathological image based on Poisson annular condition random field
Diagnostic method includes the following steps:
Step 1: acquisition digitlization cervical cancer tissues pathological image is used for systematic training, and picture format includes * .bmp, *
.BMP, * .dip, * DIP, * .jpg, * .JPG, * .jpeg, * JPEG, * .jpe, * .JPE, * .jfif, * JFIF, * .gif, *
.GIF, * .GIF, * .GIF, * .GIFf, * .GIFF, * .png, * .PNG etc.: for example, the present embodiment includes 307 height using one
In poorly differentiated cervical cancer tissues pathological image database carry out systematic training (the high, normal, basic differentiation of training set each 20, test
Collection 247), every picture size is 2560 × 1920 pixels.
Step 2: acquired image being pre-processed: histogram first being reused to image denoising using median filter
Figure equalization is to enhance picture contrast.(available grayscale image here)
It is as shown in Figure 4: step 3: the image pre-processed being split and piecemeal: using a kind of poly- based on K-means
The image partition method of class, setting cluster numbers K be 4, image gather for nucleus, cytoplasm, cytoplasm, image labeling this four
Class (only focuses on nucleus, because nucleus is the mesh of segmentation in conjunction with morphological operation cellulation core binary map appropriate
Mark).Nucleus is positioned according to image segmentation result, obtains nucleus barycentric coodinates, and the grayscale image of original image is divided into 100
The fritter (ignoring image edge pixels, image actual size is 2500 × 1900 pixels) of × 100 pixels.
Step 4: feature extraction and selection: extracting each fritter and be averaged DAISY of the DAISY feature descriptor as the fritter
Feature, i.e. local feature, then extract full figure textural characteristics (grey level histogram) and full figure is averaged DAISY feature as global special
Then sign carries out feature selecting and dimensionality reduction using Principal Component Analysis (PCA).
DAISY be towards dense characteristic extract can quickly calculate local image characteristics description son, its essential idea and
SIFT is the same: block statistics gradient orientation histogram, unlike, DAISY is improved on partition strategy, is utilized
Gaussian convolution carries out the piecemeal convergence of gradient orientation histogram, in this way using Gaussian convolution can it is quick it is computational can be fast
Speed densely carries out the extraction of Feature Descriptor.
Step 5: condition random field classifier: the feature vector or eigenmatrix input condition obtained by step 4 is random
Field model is classified, and exports classification results (differentiated, middle differentiation, low differentiation).
Model in the present embodiment is designed based on condition random field, and wherein Fig. 4 is that an X and Y has identical figure knot
The linear chain conditional random of structure, wherein open circles indicate that the variable is generated by model.
The model is a kind of undirected graph model of probability, is usually used in segmentation and flag sequence data, such as image segmentation, image point
Class, part-of-speech tagging, Chinese word segmentation etc..
Condition random field design: the layout of condition random field is protection point of the invention.
The multilayer Poisson annular layout design of condition random field is as shown below, shaped like " target ", by three circle black picture elements
(pixel and fritter mean two to the conditional probability of parameter calculating centre blue pixel (fritter) of point (or fritter) here
Person can mutual analogy, the square-shaped patterns fritter of gridding is regarded as a pixel, convenient for application designed condition with
Computer aided).
The with the following functions and advantage of this layout.
The multilayer Poisson annular design of the layout can effectively obtain the information data around target area, for example organize
Available nucleus and cytoplasm and some tissue fluid to around target area in pathological image.
The layout can increase and decrease the number of rings of " target " according to the size of real image, and larger image can suitably be increased
Add number of rings, two rings can be reduced to for smaller image, suitable for the different size histopathology as captured by different cameras
Micro-image, the histopathology micro-image after being also applied for full width histopathology micro-image and shear treatment.
The layout can be using the calculating mode of similar " scoring target practice ", and closer to central area, " score " is higher, specific gravity
It is bigger.Poisson bright spot: when monochromatic light exposure is in diameter appropriate small circular plate or ball, meeting occurs cyclic annular on optical screen later
Concentric circles each other diffraction fringe, and will appear a very small speck in the center point of all concentric circles, i.e. Poisson is bright
Spot.And the concentric circles outside our Poisson bright spots is referred to as " Poisson ring ", which is that the one kind of " Poisson ring " on the digital image is close
Like characterization.Details are referring to Fig. 6 multilayer Poisson annular layout schematic diagram.
Condition random field can indicate that wherein c indicates that class label, x are input picture with formula (1), Z (θ, x) be to point
The partition functions that cloth is normalized, θ={ θψ, θφ, θλIt is model parameter, and i is corresponded in figure in the image of grid
Index, n are full figure fritter number.
The formula explains the condition random field that the present invention designs from mathematics level, contains corresponding characteristic information.
It is as shown in Figure 7: to include the fritter of nucleus according to nucleus coordinate setting, using these fritters as layout center
Blue fritter, by 0.6 times of inner ring be averaged DAISY feature vector and 0.25 times of middle ring be averaged DAISY feature vector and
The be averaged sum of DAISY feature vector of 0.15 times of outer ring obtains the final DAISY feature vector of blue fritter.
All average DAISY feature vectors comprising nucleus fritter, the nucleus of available diagram picture are calculated again
Average DAISY feature vector (200 dimension).
According to the feature vector of available one 256 dimension of the grey level histogram of original image grayscale image.
The feature vector of one 200 dimension is obtained according to the average DAISY feature vector of original image grayscale image.
Differentiated subset, middle differentiation subset and the low differentiation subset in training set are inputted respectively, obtain 3 class models (every class
Model all includes the average value of 3 feature vectors of 20 width images).
Input test collection calculates 3 feature vectors of each image again, calculates the likelihood of each image and 3 models
(likelihood=nucleus is averaged DAISY characteristic similarity * full figure grey level histogram similarity * overall situation DAISY characteristic similarity,
Similarity uses matrix similarity calculating method, uses corr2 function), select the highest model classification of likelihood as final
Classification and diagnostic result.
Finally it should be noted that possible alternative solution is as follows:
(1) median filter method is replaced with mean filter (mean filtering), carries out image denoising.
(2) K-means clustering method is replaced with watershed (watershed) method, carries out image segmentation.
(3) DAISY feature is replaced with the constant expansible key point (BRISK) of binary robust, carries out local shape factor.
(4) grey level histogram feature is replaced with color moment (color moment) feature, carries out global characteristics extraction.
(5) Principal Component Analysis is replaced with Fisher face (LDA), carries out feature selecting.
The classification of cervical cancer tissues pathological image is in the world mostly using decision tree, support vector machines and artificial mind
Through classification methods such as networks, and it is in conceptual phase.Condition random field is made classifier applied to cervical carcinoma group in the present embodiment
The classification for knitting pathological image realizes conversion of the scientific research technology to real achievement.
In the design of conditional random field models, the present embodiment brand-new design multilayer Poisson annular layout, to combine
Using the local feature and global characteristics of cervical cancer tissues pathology micro-image, make the diagnostic result of system is highly efficient can
It leans on.
The technical principle of the invention is described above in combination with a specific embodiment, these descriptions are intended merely to explain of the invention
Principle shall not be construed in any way as a limitation of the scope of protection of the invention.Based on explaining herein, those skilled in the art
It can associate with other specific embodiments of the invention without creative labor, these modes fall within this hair
Within bright protection scope.
Claims (10)
1. a kind of cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field, which is characterized in that described
Method includes:
101, digitlization cervical cancer tissues pathological image is obtained；
102, the digitlization cervical cancer tissues pathological image of acquisition is pre-processed；
103, cluster segmentation is carried out to the pretreated digitlization cervical cancer tissues pathological image of process using partitioning algorithm and divided
Block obtains multiple images fritter；
104, feature is extracted in each image fritter obtained from step 103, feature choosing then is carried out to the feature extracted
It selects；
105, digitlization cervical cancer tissues pathological image classification is carried out to the feature chosen using conditional random field models, obtained
Obtain the classification results of cervical cancer tissues pathological image；
The conditional random field models are to be obtained using condition random field and digitlization cervical cancer tissues pathological image training
Model.
2. image diagnosing method according to claim 1, which is characterized in that
The format of the digitlization cervical cancer tissues pathological image obtained in the step 101 includes: * .bmp, * .BMP, *
.dip、*DIP、*.jpg、*.JPG、*.jpeg、*JPEG、*.jpe、*.JPE、*.jfif、*JFIF、*.gif、*.GIF、*
.GIF, * .GIF, * .GIFf, * .GIFF, * .png and * .PNG.
3. image diagnosing method according to claim 2, which is characterized in that the step 102 further includes following steps:
102A, image denoising is carried out to digitlization cervical cancer tissues pathological image using median filter；
102B, image will be enhanced by histogram equalization by the digitlization cervical cancer tissues pathological image of image denoising
Contrast.
4. image diagnosing method according to claim 3, which is characterized in that
The step 103 further includes following steps:
103A, using the image partition method clustered based on K-means to sick by pretreated digitlization cervical cancer tissues
Image of science carries out cluster segmentation；
Wherein, setting cluster numbers K is 4, image is gathered these four types of for nucleus, cytoplasm, cytoplasm, image labeling, then is tied
Close morphological operation cellulation core binary map appropriate；
103B, nucleus is positioned according to the result of image segmentation, obtains nucleus barycentric coodinates, and the grayscale image of original image point
It is segmented into the image fritter of 100 × 100 pixels.
5. image diagnosing method according to claim 4, which is characterized in that
The step 104 further includes following steps:
104A, each image fritter of extraction are averaged DAISY feature descriptor as the DAISY feature of the image fritter and are used as part
Feature；
104B, extraction full figure textural characteristics and full figure are averaged DAISY feature as global characteristics；
104C, it is handled using local feature and global characteristics of the Principal Component Analysis PCA to acquisition, obtains feature vector
Group.
6. image diagnosing method according to claim 5, which is characterized in that
The step 105 further include:
The described eigenvector group input condition random field models that selection obtains are classified, and export cervical cancer tissues pathology
Learn the classification results of image.
7. image diagnosing method according to claim 6, which is characterized in that
It is described to be indicated based on the classifier of condition random field with following formula:
Wherein, c indicates that class label, x are input picture, and Z (θ, x) is the partition functions that distribution is normalized, θ={ θψ,
θφ, θλ) it is model parameter, and i corresponds to the index in figure in the image of grid, n is full graphics image fritter number.
8. image diagnosing method according to claim 5, which is characterized in that
Described eigenvector group includes: that the average DAISY feature vector of full figure nucleus, the overall situation of full figure are averaged DAISY feature
The grey level histogram vector of the full grayscale image of vector sum.
9. image diagnosing method according to claim 1, which is characterized in that
When being pre-processed in the step 102 to the digitlization cervical cancer tissues pathological image of acquisition, using mean filter
Method carries out image denoising to the digitlization cervical cancer tissues pathological image.
The partitioning algorithm used in the step 103 is watershed algorithm；
Feature choosing is carried out to the feature extracted using principal component analytical method or Fisher face in the step 103
It selects.
10. the image diagnosing method according to requiring 8, which is characterized in that
The conditional random field models include: differentiated model, middle differentiation model and low differentiation model；
The differentiated model is the average value of the feature vector group of the digitlization cervical cancer tissues pathological image of differentiated；
The middle differentiation model is the average value of the feature vector group of the digitlization cervical cancer tissues pathological image of middle differentiation；
The low differentiation model is the average value of the feature vector group of poorly differentiated digitlization cervical cancer tissues pathological image；
The conditional random field models can be according to the feature vector component of the digitlization cervical cancer tissues pathological image of input
Likelihood is not calculated with the differentiated model, middle differentiation model and low differentiation model, finally chooses the highest model of likelihood
Classification is as final classification results；
The classification results include: differentiated, middle differentiation and low differentiation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811242279.9A CN109410196A (en) | 2018-10-24 | 2018-10-24 | Cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811242279.9A CN109410196A (en) | 2018-10-24 | 2018-10-24 | Cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109410196A true CN109410196A (en) | 2019-03-01 |
Family
ID=65469052
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811242279.9A Pending CN109410196A (en) | 2018-10-24 | 2018-10-24 | Cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109410196A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993740A (en) * | 2019-04-02 | 2019-07-09 | 广州中医药大学(广州中医药研究院) | A kind of cell cancer image grading method and device |
| CN110264454A (en) * | 2019-06-19 | 2019-09-20 | 四川智动木牛智能科技有限公司 | Cervical cancer tissues pathological image diagnostic method based on more hidden layer condition random fields |
| CN110349170A (en) * | 2019-07-13 | 2019-10-18 | 长春工业大学 | A kind of full connection CRF cascade FCN and K mean value brain tumor partitioning algorithm |
| CN110427845A (en) * | 2019-07-19 | 2019-11-08 | 东莞弓叶互联科技有限公司 | Article pixel center determines method, apparatus, equipment and readable storage medium storing program for executing |
| CN110852396A (en) * | 2019-11-15 | 2020-02-28 | 苏州中科华影健康科技有限公司 | Sample data processing method for cervical image |
| CN112334990A (en) * | 2019-06-04 | 2021-02-05 | 艾多特公司 | Automatic cervical cancer diagnosis system |
| WO2021164550A1 (en) * | 2020-02-18 | 2021-08-26 | Oppo广东移动通信有限公司 | Image classification method and apparatus |
| CN114565611A (en) * | 2022-04-28 | 2022-05-31 | 武汉大学 | Medical information acquisition method and related equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103096786A (en) * | 2010-05-03 | 2013-05-08 | 国际科学技术医疗系统有限责任公司 | Image analysis for cervical neoplasia detection and diagnosis |
| CN106570505A (en) * | 2016-11-01 | 2017-04-19 | 北京昆仑医云科技有限公司 | Method for analyzing histopathologic image and system thereof |
| CN106600571A (en) * | 2016-11-07 | 2017-04-26 | 中国科学院自动化研究所 | Brain tumor automatic segmentation method through fusion of full convolutional neural network and conditional random field |
| US20170277981A1 (en) * | 2016-01-08 | 2017-09-28 | Siemens Healthcare Gmbh | Deep Image-to-Image Network Learning for Medical Image Analysis |
-
2018
- 2018-10-24 CN CN201811242279.9A patent/CN109410196A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103096786A (en) * | 2010-05-03 | 2013-05-08 | 国际科学技术医疗系统有限责任公司 | Image analysis for cervical neoplasia detection and diagnosis |
| US20170277981A1 (en) * | 2016-01-08 | 2017-09-28 | Siemens Healthcare Gmbh | Deep Image-to-Image Network Learning for Medical Image Analysis |
| CN106570505A (en) * | 2016-11-01 | 2017-04-19 | 北京昆仑医云科技有限公司 | Method for analyzing histopathologic image and system thereof |
| CN106600571A (en) * | 2016-11-07 | 2017-04-26 | 中国科学院自动化研究所 | Brain tumor automatic segmentation method through fusion of full convolutional neural network and conditional random field |
Non-Patent Citations (2)
| Title |
|---|
| YIMING LIU 等: "Automatic Segmentation of Cervical Nuclei Based on Deep Learning and a Conditional Random Field", 《IEEE》 * |
| 阳维 等: "基于图像块分类器和条件随机场的显微图像分割", 《计算机应用》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993740A (en) * | 2019-04-02 | 2019-07-09 | 广州中医药大学(广州中医药研究院) | A kind of cell cancer image grading method and device |
| CN112334990A (en) * | 2019-06-04 | 2021-02-05 | 艾多特公司 | Automatic cervical cancer diagnosis system |
| CN110264454A (en) * | 2019-06-19 | 2019-09-20 | 四川智动木牛智能科技有限公司 | Cervical cancer tissues pathological image diagnostic method based on more hidden layer condition random fields |
| CN110349170A (en) * | 2019-07-13 | 2019-10-18 | 长春工业大学 | A kind of full connection CRF cascade FCN and K mean value brain tumor partitioning algorithm |
| CN110349170B (en) * | 2019-07-13 | 2022-07-08 | 长春工业大学 | Full-connection CRF cascade FCN and K mean brain tumor segmentation algorithm |
| CN110427845A (en) * | 2019-07-19 | 2019-11-08 | 东莞弓叶互联科技有限公司 | Article pixel center determines method, apparatus, equipment and readable storage medium storing program for executing |
| CN110427845B (en) * | 2019-07-19 | 2022-12-16 | 广东弓叶科技有限公司 | Method, device and equipment for determining pixel center of article and readable storage medium |
| CN110852396A (en) * | 2019-11-15 | 2020-02-28 | 苏州中科华影健康科技有限公司 | Sample data processing method for cervical image |
| WO2021164550A1 (en) * | 2020-02-18 | 2021-08-26 | Oppo广东移动通信有限公司 | Image classification method and apparatus |
| CN114565611A (en) * | 2022-04-28 | 2022-05-31 | 武汉大学 | Medical information acquisition method and related equipment |
| CN114565611B (en) * | 2022-04-28 | 2022-07-19 | 武汉大学 | Medical information acquisition method and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109410196A (en) | Cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field | |
| Silva-Rodríguez et al. | Going deeper through the Gleason scoring scale: An automatic end-to-end system for histology prostate grading and cribriform pattern detection | |
| CN109389129B (en) | Image processing method, electronic device and storage medium | |
| Beevi et al. | Automatic mitosis detection in breast histopathology images using convolutional neural network based deep transfer learning | |
| Irshad et al. | Methods for nuclei detection, segmentation, and classification in digital histopathology: a review—current status and future potential | |
| Telrandhe et al. | Detection of brain tumor from MRI images by using segmentation & SVM | |
| Zhang et al. | Automated semantic segmentation of red blood cells for sickle cell disease | |
| CN111931811B (en) | Calculation method based on super-pixel image similarity | |
| Pan et al. | Mitosis detection techniques in H&E stained breast cancer pathological images: A comprehensive review | |
| US11954593B2 (en) | Method to determine a degree of abnormality, a respective computer readable medium and a distributed cancer analysis system | |
| CN109299679A (en) | Cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field | |
| Di Ruberto et al. | A leucocytes count system from blood smear images: Segmentation and counting of white blood cells based on learning by sampling | |
| Wang et al. | Detection of dendritic spines using wavelet‐based conditional symmetric analysis and regularized morphological shared‐weight neural networks | |
| CN110264454A (en) | Cervical cancer tissues pathological image diagnostic method based on more hidden layer condition random fields | |
| Mercan et al. | Deep feature representations for variable-sized regions of interest in breast histopathology | |
| Phonsa et al. | A survey: image segmentation techniques | |
| Rampun et al. | Breast density classification using local ternary patterns in mammograms | |
| Vajravelu et al. | Machine learning techniques to detect bleeding frame and area in wireless capsule endoscopy video | |
| Madduri et al. | Classification of breast cancer histopathological images using convolutional neural networks | |
| Zhou et al. | Semantic instance segmentation with discriminative deep supervision for medical images | |
| Zou et al. | Deep learning for nuclei segmentation and cell classification in cervical liquid based cytology | |
| Jiang et al. | Parallel ‘same’and ‘valid’convolutional block and input-collaboration strategy for histopathological image classification | |
| Pan et al. | A review of machine learning approaches, challenges and prospects for computational tumor pathology | |
| Avalos et al. | An accurate Cluster chaotic optimization approach for digital medical image segmentation | |
| Valliappan et al. | A theoretical methodology and prototype implementation for detection segmentation classification of digital mammogram tumor by machine learning and problem solving approach |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190301 |
|
| RJ01 | Rejection of invention patent application after publication |