CN109299679A - Cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field - Google Patents
Cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field Download PDFInfo
- Publication number
- CN109299679A CN109299679A CN201811056305.9A CN201811056305A CN109299679A CN 109299679 A CN109299679 A CN 109299679A CN 201811056305 A CN201811056305 A CN 201811056305A CN 109299679 A CN109299679 A CN 109299679A
- Authority
- CN
- China
- Prior art keywords
- image
- cervical cancer
- cancer tissues
- random field
- 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
- 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
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- 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
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/695—Preprocessing, e.g. image segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
-
- 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
- G06T2207/10056—Microscopic 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
- G06T2207/20081—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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Probability & Statistics with Applications (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to disease identification technical field more particularly to a kind of cervical cancer tissues pathological image diagnostic methods based on sleeve configuration condition random field.This method comprises the following steps: A1, to cervical cancer tissues pathology micro-image to be processed, is pre-processed;A2, processing is split to pretreated cervical cancer tissues pathology micro-image;A3, the image progress feature extraction after dividing processing is chosen, obtains the global characteristics of extraction;A4, the image after global characteristics, dividing processing is handled by the way of condition random field, obtains feature vector;In A5, the disaggregated model for training described eigenvector input in advance, the classification results of cervical cancer tissues pathology micro-image are obtained.Method brand-new design of the invention sleeve configuration central symmetry layout, the local feature and global characteristics of cervical cancer tissues pathology micro-image is used in combination, makes the diagnostic result more high efficient and reliable of system.
Description
Technical field
The invention belongs to disease identification technical field more particularly to a kind of cervical carcinoma groups based on sleeve configuration condition random field
Knit pathological image diagnostic method.
Background technique
Cervical carcinoma is to lead to the most common malignant tumour of second of female patient death, and disease incidence is only second to mammary gland
Cancer.Cervical carcinoma is the specific malignant tumour of unique cause of disease in the world, high-risk HPV persistent infection be cause cervical carcinoma it is main because
Element.
Currently, using histopathology image as a basic fact in the screening process of cervical carcinoma, use condition is random
Field classifies to cervicovaginal mirror image, includes the following steps: (1), pre-processes to image, including image calibration, figure
As being registrated and dissecting feature extraction three parts;(2), based on K mean cluster method by image segmentation, identifying in color and intensity is
Subregion in each organization type of homogeneity;(3), diagnosis includes but is not limited to Acetowhitening, inlays, dotted and atypia blood vessel
Correlated characteristic;(4), the classifier of the CRF model based on the classification results according to probabilistic manner combination adjacent area: n is utilized
A different characteristic f1, f2, f3 ... the characteristic function (Y1, Y2, Y3 and Y4) of fn have scaly epithelium, columnar epithelium, zone of transformation
With four kinds of cluster situations (W1, W2, W3 and W4) of four kinds of histological types of cervix opening, from K arest neighbors (KNN) classifier and
All results of linear differential analysis (LDA) classifier determine the conditional probability distribution (it is assumed that only diagnostic characteristic), using most
Big posteriority (MAP) estimates the parameter to determine Posterior probability distribution;(5) detection is determined in the way of by new based on window and is examined
The sensitivity and specificity of disconnected algorithm, obtain diagnostic result.
However, cervical cancer tissues pathological image is intended only as basic fact, it is still necessary to which experience is rich in this screening process
Rich virologist judges image, but different pathological scholar or the same virologist of different time are to same disease
The judgement for managing image is also variant, and there may be biggish errors for this;Meanwhile the limited amount of virologist, and it is less-developed
Regional healthcare resource shortage, virologist are equally rare;Experience medico not abundant enough or virologist, to histopathology
Image cannot make reliable judgement.
In this screening process, it is only capable of judging whether regional area is different using a classifier based on condition random field
Often, the classification results of cervical carcinoma, such as malignant tumour (high, normal, basic differentiation three-level) can not be provided.
Summary of the invention
(1) technical problems to be solved
For existing technical problem, the present invention provides a kind of cervical cancer tissues based on sleeve configuration condition random field
Pathological image diagnostic method, this method brand-new design sleeve configuration central symmetry layout, cervical cancer tissues disease is used in combination
The local feature and global characteristics of micro-image of science make the diagnostic result more high efficient and reliable of system.
(2) technical solution
In order to achieve the above object, the main technical schemes that the present invention uses include:
A kind of cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field, includes the following steps:
A1, to cervical cancer tissues pathology micro-image to be processed, pre-processed;
A2, processing is split to pretreated cervical cancer tissues pathology micro-image;
A3, the image progress feature extraction after dividing processing is chosen, obtains the global characteristics of extraction;
A4, the image after global characteristics, dividing processing is handled by the way of condition random field, obtain feature to
Amount;
In A5, the disaggregated model for training described eigenvector input in advance, cervical cancer tissues pathology micrograph is obtained
The classification results of picture;
Wherein, the disaggregated model be based on be pre-established in the training image database of classification results using condition with
The model of the mode training on airport;
In the training process of disaggregated model, the Partial Feature of each training image is obtained by the way of condition random field
Vector, that is, select for the pixel of each training image, obtained using the pixel selection mode that sleeve configuration central symmetry is laid out
The conditional probability for the central pixel point for taking sleeve configuration central symmetry to be laid out, to obtain the Partial Feature vector of the training image.
Further, condition random field can be used following formula to indicate:
Wherein c presentation class resulting class label, x are the training image of input, and Z (θ, x) is to carry out normalizing to distribution
The partition functions of change, θ={ θψ, θφ, θλBe disaggregated model model parameter, i be for each training image in dividing processing
The index of each fritter afterwards, n are full figure pixel number.
Further, in the step A1, pretreatment includes using median filter or Gaussian filter to image denoising;
Using the contrast of histogram equalization enhancing image, grayscale image is obtained.
Further, in the step A2, using a kind of image partition method handle clustered based on level set or K-means
Grayscale image gather for nucleus, cytoplasm, cytoplasm, image labeling it is these four types of, then by nucleus one kind combining form therein
Learn operation cellulation core binary map;
It is the fritter or pixel of 100 × 100 pixels the grayscale image piecemeal in step A1.
Further, in step A3, the average DAISY feature descriptor of each fritter is extracted as the fritter
DAISY feature, i.e. local feature;
The textural characteristics of grayscale image full figure and the average DAISY feature of grayscale image full figure are extracted again as global characteristics;
The textural characteristics include grey level histogram feature or color histogram feature;
Feature selecting and dimensionality reduction are carried out to global characteristics using Principal Component Analysis or Fisher face, obtain full figure
Grey level histogram feature vector and full figure are averaged DAISY feature vector.
Further, in step A4, nucleus is positioned according to nucleus binary map, includes cell according to apoptotic nueleolus
The fritter of core, the central pixel point that these fritters are laid out respectively as sleeve configuration central symmetry;
It according to central pixel point, is laid out by sleeve configuration central symmetry, determines inside and outside the two of sleeve configuration central symmetry layout
Pixel is enclosed, nucleus is obtained according to the local feature of each pixel and is averaged DAISY feature vector.
Further, be averaged DAISY feature vector, full figure grey level histogram feature vector and full figure of nucleus is averaged
Feature vector of the DAISY feature vector as input disaggregated model.
Further, the training image database includes differentiated subset, middle differentiation subset and low differentiation subset image,
It include 20 images in each subset.
Further, the size of every described image is 2560 × 1920 pixels.
(3) beneficial effect
The beneficial effects of the present invention are:
1, condition random field is made classifier applied to palace by cervical cancer tissues pathological image diagnostic method provided by the invention
The classification of neck cancer histopathology image, brand-new design sleeve configuration central symmetry layout, cervical cancer tissues are used in combination
The local feature and global characteristics of pathology micro-image make the diagnostic result more high efficient and reliable of system.
2, cervical cancer tissues pathological image diagnostic method provided by the invention can accurately provide the pathology shape of patient
3 classifications such as state, including differentiated, middle differentiation, low differentiation, improve accuracy rate of diagnosis.
3, cervical cancer tissues pathological image diagnostic method provided by the invention can suitably slow down virologist's resource allocation
Uneven, the problem of being not enough, medico can also be instructed rationally to be diagnosed with the insufficient doctor of experience.
4, cervical cancer tissues pathological image diagnostic method provided by the invention, can solve between different doctors and same doctor
Contradictory problems under different conditions give patient one more reliable diagnostic result.
Detailed description of the invention
Fig. 1 is the schematic diagram of linear chain conditional random in the embodiment of the present invention;
Fig. 2 is the sleeve configuration central symmetry schematic layout pattern of conditional of embodiment of the present invention random field;
Fig. 3 is the analysis flow chart diagram of the embodiment of the present invention;
Fig. 4 is the grayscale image of cervical carcinoma histopathology image in the embodiment of the present invention;
Fig. 5 is the block diagram of cervical carcinoma histopathology image grayscale figure in the embodiment of the present invention.
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.
Disaggregated model is constructed in accordance with the following steps:
307 S1, acquisition cervical cancer tissues pathology micro-images (hereinafter referred to as image) are used as training image data
Library is used for systematic training.
The training image database includes training set and test set, wherein include in training set differentiated subset, in point
Beggar's collection and each 20 of low differentiation subset, test set 247 is opened.Every picture size is 2560 × 1920 pixels, picture format packet
Include * .bmp, * .BMP, * .dip, * DIP, * .jpg, * .JPG, * .jpeg, * JPEG, * .jpe, * .JPE, * .jfif, * JFIF, *
.gif, * .GIF, * .GIF, * .GIF, * .GIFf, * .GIFF, * .png, * .PNG etc..
S2, acquired image is pre-processed: it is equal that histogram first is reused to image denoising using median filter
Weighing apparatusization enhances picture contrast, obtains the grayscale image of image.
Wherein, Gaussian filter also can be used instead of median filter, to image denoising.
S3, grayscale image is split: using a kind of image partition method based on K-means cluster, cluster numbers K is set
It is 4, grayscale image is gathered these four types of for nucleus, cytoplasm, cytoplasm and image labeling, then will wherein nucleus one kind combines
Morphological operation cellulation core binary map.Grayscale image is divided into the fritter (pixel) of 100 × 100 pixels.
Wherein, ignore grayscale image edge pixel, actual size is 2500 × 1900 pixels.
Wherein, level set (level set) method also can be used to cluster instead of K-means, carry out image segmentation.
S4, the image progress feature extraction after dividing processing is chosen, obtains the global characteristics of extraction: it is flat extracts each fritter
Equal DAISY feature of the DAISY feature descriptor as the fritter, i.e. local feature;The textural characteristics of grayscale image full figure are extracted again
(being full figure grey level histogram feature in present embodiment) and full figure are averaged DAISY feature as global characteristics;Then using master
Componential analysis (PCA) carries out feature selecting and dimensionality reduction, obtains full figure grey level histogram feature vector B and full figure is averaged DAISY
Feature vector C.
Wherein, in local shape factor, Scale invariant features transform (SIFT) also can be used instead of DAISY feature.
DAISY is local image characteristics description that can quickly calculate extracted towards dense characteristic, and essential idea and SIFT are one
Sample: block statistics gradient orientation histogram, unlike, DAISY is improved on partition strategy, utilizes Gaussian convolution
Carry out the piecemeal convergence of gradient orientation histogram, in this way using Gaussian convolution can it is quick it is computational can quickly densely
Carry out the extraction of Feature Descriptor.
It can infer, color histogram feature also can be used instead of grey level histogram feature, extract color histogram
Feature and full figure are averaged DAISY feature as global characteristics.
It can infer, Fisher face (LDA) also can be used instead of Principal Component Analysis (PCA), carry out special
Sign selection and dimensionality reduction.
S5, feature vector one condition random field of input is classified, constructs disaggregated model.
As shown in Figure 1, there is the linear chain conditional random of identical graph structure for an X and Y, wherein open circles indicate to be somebody's turn to do
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 schemes
As segmentation, image classification, part-of-speech tagging, Chinese word segmentation etc..
Condition random field can be used following formula to indicate:
Wherein c presentation class resulting class label, x are the training image of input, and Z (θ, x) is to carry out normalizing to distribution
The partition functions of change, θ={ θψ, θφ, θλBe disaggregated model model parameter, i be for each training image in dividing processing
The index of each fritter afterwards, n are full figure pixel number.
Condition random field uses sleeve configuration central symmetry layout designs as shown in Fig. 2, shaped like " eyes ", by outer ring and interior
The parameter for enclosing pixel calculates the conditional probability of central pixel point, and wherein inner ring pixel is closer to central pixel point.
The with the following functions and advantage of this layout: (1), the layout has during inside and outside two circle pixel (fritter) calculates jointly
The conditional probability of imago vegetarian refreshments (fritter), it is possible to reduce error caused by accidentalia on a certain circle, such as a certain picture of inner ring
There is poorly differentiated cervical cancer cell in vegetarian refreshments, and other pixels are differentiated cervical cancer cell, is calculated in this way
The conditional probability of central pixel point just substantially conforms to differentiated cervical cancer cell;(2), the layout can be internally on outermost ring of pixels point
Conditional probability different weights is set, keep result more acurrate, such as histopathology micro-image, inner ring pixel pair
The influence of central pixel point is bigger, can be set to high weight, and influence of the outermost ring of pixels point to central pixel point is small, can be set
For low weight;(3), the shape of the layout can satisfy a certain range of most cells around trap center point nucleus
Core, there are also partial cytoplasms and cytoplasm, this is highly beneficial for histopathology image analysis.
The center of gravity and its coordinate of each nucleus are calculated according to nucleus binary map, nucleus is positioned, according to cell
Core coordinate setting includes the fritter of nucleus, as target area, using these fritters as the central pixel point at layout center, root
Outer ring and inner ring pixel are obtained according to sleeve configuration central symmetry layout designs as shown in Figure 2.It is put down by 0.6 times of inner ring pixel
The be averaged sum of DAISY feature vector of equal DAISY feature vector and 0.4 times of outermost ring of pixels point obtains the final of central pixel point
DAISY feature vector, then all average DAISY feature vectors comprising nucleus fritter are calculated, available 1 200 dimension
Nucleus is averaged DAISY feature vector A.
According to the feature vector B of available 1 256 dimension of full figure grey level histogram feature in step S4.
According to full figure in step S4 be averaged DAISY feature obtain 1 200 dimension feature vector C.
Above 3 feature vectors are inputted to differentiated subset, middle differentiation subset and the low differentiation subset in training set respectively,
3 class model of differentiated model, middle differentiation model and low differentiation model is obtained, every class model all includes 3 features of 20 images
The average value of vector.The image that input test is concentrated again calculates 3 feature vectors of every image, calculates similar to 3 class models
Rate selects the highest model classification of likelihood as final classification and diagnostic result (differentiated, middle differentiation, low differentiation).
Wherein, likelihood uses matrix likelihood calculation method, and formula is as follows:
Likelihood=corr2 (A, AIt is high/medium/low)*corr2(B,BIt is high/medium/low)*corr2(C,CIt is high/medium/low)。
Likelihood=nucleus DAISY feature likelihood * full figure grey level histogram likelihood * full figure that is averaged is averaged DAISY
Feature likelihood.
As shown in figure 3, providing a kind of cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field, have
Gymnastics is made as follows:
A1, to cervical cancer tissues pathology micro-image to be processed, pre-processed: first using median filter to figure
As denoising, histogram equalization is reused to enhance picture contrast, obtains the grayscale image of image, as shown in Figure 4.Wherein,
Gaussian filter can be used instead of median filter, to image denoising.
A2, processing is split to pretreated cervical cancer tissues pathology micro-image: using a kind of based on K-means
The image partition method of cluster, setting cluster numbers K are 4, and grayscale image is gathered for nucleus, cytoplasm, cytoplasm and image mark
Note is these four types of, then wherein nucleus one kind combining form will operate cellulation core binary map.Grayscale image piecemeal be 100 ×
The fritter (pixel) of 100 pixels, as shown in Figure 5.
Wherein, ignore grayscale image edge pixel, actual size is 2500 × 1900 pixels.
Wherein, level set (level set) method also can be used to cluster instead of K-means, carry out image segmentation.
A3, the image progress feature extraction after dividing processing is chosen, obtains the global characteristics of extraction: it is flat extracts each fritter
Equal DAISY feature of the DAISY feature descriptor as the pixel, i.e. local feature;The texture for extracting grayscale image full figure again is special
Sign (being full figure grey level histogram feature in present embodiment) and full figure are averaged DAISY feature as global characteristics;Then it uses
Principal Component Analysis (PCA) carries out feature selecting and dimensionality reduction, obtains full figure grey level histogram feature vector and full figure is averaged DAISY
Feature vector.
Wherein, in local shape factor, Scale invariant features transform (SIFT) also can be used instead of DAISY feature.
DAISY is local image characteristics description that can quickly calculate extracted towards dense characteristic, and essential idea and SIFT are one
Sample: block statistics gradient orientation histogram, unlike, DAISY is improved on partition strategy, utilizes Gaussian convolution
Carry out the piecemeal convergence of gradient orientation histogram, in this way using Gaussian convolution can it is quick it is computational can quickly densely
Carry out the extraction of Feature Descriptor.
It can infer, color histogram feature also can be used instead of grey level histogram feature, extract color histogram
Feature and full figure are averaged DAISY feature as global characteristics.
It can infer, Fisher face (LDA) also can be used instead of Principal Component Analysis (PCA), carry out special
Sign selection and dimensionality reduction.
A4, the image after global characteristics, dividing processing is handled by the way of condition random field, obtain feature to
Amount: the center of gravity and its coordinate of each nucleus being calculated according to nucleus binary map, nucleus are positioned, according to nucleus coordinate
Positioning includes the fritter of nucleus, these fritters respectively as the central pixel point at layout center, according to as shown in Figure 2 narrow
Elongated central symmetry layout designs obtain outer ring and inner ring pixel.It is averaged DAISY feature vector by 0.6 times of inner ring pixel
The final DAISY feature vector of central pixel point is obtained with the be averaged sum of DAISY feature vector of 0.4 times of outermost ring of pixels point, then
Calculate all average DAISY feature vectors comprising nucleus pixel, the nucleus of available 1 200 dimension is averaged DAISY
Feature vector.
A5, DAISY feature vector that nucleus is averaged, full figure grey level histogram feature vector and full figure be averaged DAISY spy
It levies in vector input disaggregated model trained in advance, obtains the classification results of cervical carcinoma micro-image.
In the design of conditional random field models, brand-new design of the present invention sleeve configuration central symmetry layout, to combine
Using the local feature and global characteristics of cervical cancer tissues pathology micro-image, the diagnostic result of system is made more efficiently may be used
It leans on.
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 applied to cervical cancer tissues pathological image by the present invention
It automatically analyzes, forms the auto-check system of a set of cervical carcinoma micro-image, realize scientific research technology and turn to real achievement
Change.
Meanwhile the problem of can suitably slowing down virologist's resource allocation unevenness, being not enough, medico can also be instructed
Insufficient doctor is rationally diagnosed with experience;The contradictory problems between different doctors and under same doctor's different conditions are solved,
To patient one more reliable diagnostic result;Accurately provide the pathological state of patient, including differentiated, middle differentiation, low point
3 classifications changed improve accuracy rate of diagnosis.
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 (9)
1. a kind of cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field, which is characterized in that including such as
Lower step:
A1, to cervical cancer tissues pathology micro-image to be processed, pre-processed;
A2, processing is split to pretreated cervical cancer tissues pathology micro-image;
A3, the image progress feature extraction after dividing processing is chosen, obtains the global characteristics of extraction;
A4, the image after global characteristics, dividing processing is handled by the way of condition random field, obtains feature vector;
In A5, the disaggregated model for training described eigenvector input in advance, cervical cancer tissues pathology micro-image is obtained
Classification results;
Wherein, the disaggregated model is based on being pre-established in the training image database of classification results using condition random field
Mode training model;
In the training process of disaggregated model, obtained by the way of condition random field the Partial Feature of each training image to
Amount, that is, select for the pixel of each training image, obtained using the pixel selection mode of sleeve configuration central symmetry layout
The conditional probability of the central pixel point of sleeve configuration central symmetry layout, to obtain the Partial Feature vector of the training image.
2. the cervical cancer tissues pathological image diagnostic method according to claim 1 based on sleeve configuration condition random field,
It is characterized in that, condition random field can be used following formula to indicate:
Wherein c presentation class resulting class label, x are the training image of input, and Z (θ, x) is that distribution is normalized
Partition functions, θ={ θψ, θφ, θλBe disaggregated model model parameter, i is every after dividing processing for each training image
The index of one fritter, n are full figure pixel number.
3. the cervical cancer tissues pathological image diagnostic method according to claim 1 based on sleeve configuration condition random field,
It is characterized in that, in the step A1, pretreatment includes using median filter or Gaussian filter to image denoising;
Using the contrast of histogram equalization enhancing image, grayscale image is obtained.
4. the cervical cancer tissues pathological image diagnostic method according to claim 3 based on sleeve configuration condition random field,
It is characterized in that, in the step A2, grayscale image is gathered using a kind of image partition method clustered based on level set or K-means
It is these four types of for nucleus, cytoplasm, cytoplasm, image labeling, then nucleus one kind combining form therein is operated and is given birth to
At nucleus binary map;
It is the fritter or pixel of 100 × 100 pixels the grayscale image piecemeal in step A1.
5. the cervical cancer tissues pathological image diagnostic method according to claim 4 based on sleeve configuration condition random field,
It is characterized in that, in step A3, extracts DAISY feature of the average DAISY feature descriptor of each fritter as the fritter, i.e.,
Local feature;
The textural characteristics of grayscale image full figure and the average DAISY feature of grayscale image full figure are extracted again as global characteristics;
The textural characteristics include grey level histogram feature or color histogram feature;
Feature selecting and dimensionality reduction are carried out to global characteristics using Principal Component Analysis or Fisher face, obtain full figure gray scale
Histogram feature vector sum full figure is averaged DAISY feature vector.
6. the cervical cancer tissues pathological image diagnostic method according to claim 5 based on sleeve configuration condition random field,
It is characterized in that, in step A4, nucleus is positioned according to nucleus binary map, includes the small of nucleus according to apoptotic nueleolus
Block, the central pixel point that these fritters are laid out respectively as sleeve configuration central symmetry;
It according to central pixel point, is laid out by sleeve configuration central symmetry, determines the inside and outside two circles picture of sleeve configuration central symmetry layout
Vegetarian refreshments obtains nucleus according to the local feature of each pixel and is averaged DAISY feature vector.
7. the cervical cancer tissues pathological image diagnostic method according to claim 6 based on sleeve configuration condition random field,
It is characterized in that, be averaged DAISY feature vector, full figure grey level histogram feature vector and full figure of nucleus is averaged DAISY feature
Feature vector of the vector as input disaggregated model.
8. the cervical cancer tissues pathological image diagnostic method according to claim 1 based on sleeve configuration condition random field,
It is characterized in that, the training image database includes differentiated subset, middle differentiation subset and low differentiation subset image, each subset
In include 20 images.
9. the cervical cancer tissues pathological image diagnostic method according to claim 8 based on sleeve configuration condition random field,
It is characterized in that, the size of every described image is 2560 × 1920 pixels.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811056305.9A CN109299679A (en) | 2018-09-11 | 2018-09-11 | Cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811056305.9A CN109299679A (en) | 2018-09-11 | 2018-09-11 | Cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109299679A true CN109299679A (en) | 2019-02-01 |
Family
ID=65166488
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811056305.9A Pending CN109299679A (en) | 2018-09-11 | 2018-09-11 | Cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109299679A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110136113A (en) * | 2019-05-14 | 2019-08-16 | 湖南大学 | A kind of vagina pathology image classification method based on convolutional neural networks |
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 |
CN110415246A (en) * | 2019-08-06 | 2019-11-05 | 东北大学 | A kind of analysis method of stomach fat ingredient |
WO2021080007A1 (en) * | 2019-10-23 | 2021-04-29 | 国立大学法人大阪大学 | Cancer assessment device, cancer assessment method, and program |
CN112884737A (en) * | 2021-02-08 | 2021-06-01 | 武汉大学 | Automatic mitosis detection method in breast cancer pathological image based on multistage iteration |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392228A (en) * | 2014-12-19 | 2015-03-04 | 中国人民解放军国防科学技术大学 | Unmanned aerial vehicle image target class detection method based on conditional random field model |
CN108364288A (en) * | 2018-03-01 | 2018-08-03 | 北京航空航天大学 | Dividing method and device for breast cancer pathological image |
US20180253591A1 (en) * | 2017-03-03 | 2018-09-06 | Case Western Reserve University | Predicting cancer progression using cell run length features |
-
2018
- 2018-09-11 CN CN201811056305.9A patent/CN109299679A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392228A (en) * | 2014-12-19 | 2015-03-04 | 中国人民解放军国防科学技术大学 | Unmanned aerial vehicle image target class detection method based on conditional random field model |
US20180253591A1 (en) * | 2017-03-03 | 2018-09-06 | Case Western Reserve University | Predicting cancer progression using cell run length features |
CN108364288A (en) * | 2018-03-01 | 2018-08-03 | 北京航空航天大学 | Dividing method and device for breast cancer pathological image |
Non-Patent Citations (2)
Title |
---|
JOHN LAFFERTY 等: "Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data", 《INTERNATIONAL CONFERENCE ON MACHINE LEARNING 2001》 * |
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 (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110136113A (en) * | 2019-05-14 | 2019-08-16 | 湖南大学 | A kind of vagina pathology image classification method based on convolutional neural networks |
CN110136113B (en) * | 2019-05-14 | 2022-06-07 | 湖南大学 | Vagina pathology image classification method based on convolutional neural network |
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 |
CN110264454B (en) * | 2019-06-19 | 2021-07-30 | 四川智动木牛智能科技有限公司 | Cervical cancer histopathological image diagnosis method based on multi-hidden-layer conditional random field |
CN110415246A (en) * | 2019-08-06 | 2019-11-05 | 东北大学 | A kind of analysis method of stomach fat ingredient |
WO2021080007A1 (en) * | 2019-10-23 | 2021-04-29 | 国立大学法人大阪大学 | Cancer assessment device, cancer assessment method, and program |
CN112884737A (en) * | 2021-02-08 | 2021-06-01 | 武汉大学 | Automatic mitosis detection method in breast cancer pathological image based on multistage iteration |
CN112884737B (en) * | 2021-02-08 | 2022-07-19 | 武汉大学 | Automatic mitosis detection method in breast cancer pathological image based on multistage iteration |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | DeepPap: deep convolutional networks for cervical cell classification | |
Sari et al. | Unsupervised feature extraction via deep learning for histopathological classification of colon tissue images | |
Jia et al. | Detection of cervical cancer cells based on strong feature CNN-SVM network | |
US10235755B2 (en) | High-throughput adaptive sampling for whole-slide histopathology image analysis | |
Angel Arul Jothi et al. | A survey on automated cancer diagnosis from histopathology images | |
CN109299679A (en) | Cervical cancer tissues pathological image diagnostic method based on sleeve configuration condition random field | |
Wan et al. | Automated grading of breast cancer histopathology using cascaded ensemble with combination of multi-level image features | |
Beevi et al. | A multi-classifier system for automatic mitosis detection in breast histopathology images using deep belief networks | |
US8139831B2 (en) | System and method for unsupervised detection and gleason grading of prostate cancer whole mounts using NIR fluorscence | |
CN109034045A (en) | A kind of leucocyte automatic identifying method based on convolutional neural networks | |
Mi et al. | Deep learning-based multi-class classification of breast digital pathology images | |
CN109410196A (en) | Cervical cancer tissues pathological image diagnostic method based on Poisson annular condition random field | |
CN109658411A (en) | A kind of correlation analysis based on CT images feature Yu Patients with Non-small-cell Lung prognosis situation | |
Hu et al. | Classification of metaphase chromosomes using deep convolutional neural network | |
US11954593B2 (en) | Method to determine a degree of abnormality, a respective computer readable medium and a distributed cancer analysis system | |
CN109635846A (en) | A kind of multiclass medical image judgment method and system | |
Pan et al. | Mitosis detection techniques in H&E stained breast cancer pathological images: A comprehensive review | |
Nateghi et al. | Maximized inter-class weighted mean for fast and accurate mitosis cells detection in breast cancer histopathology images | |
CN110264454A (en) | Cervical cancer tissues pathological image diagnostic method based on more hidden layer condition random fields | |
CN109671072A (en) | Cervical cancer tissues pathological image diagnostic method based on spotted arrays condition random field | |
Clocksin | Automatic segmentation of overlapping nuclei with high background variation using robust estimation and flexible contour models | |
Abdelhafeez et al. | Skin Cancer Detection using Neutrosophic c-means and Fuzzy c-means Clustering Algorithms | |
Madduri et al. | Classification of breast cancer histopathological images using convolutional neural networks | |
Song et al. | Automated segmentation of overlapping cytoplasm in cervical smear images via contour fragments | |
Nelson et al. | An effective approach for the nuclei segmentation from breast histopathological images using star-convex polygon |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190201 |