CN107730499A - A kind of leucocyte classification method based on nu SVMs - Google Patents
A kind of leucocyte classification method based on nu SVMs Download PDFInfo
- Publication number
- CN107730499A CN107730499A CN201711048945.0A CN201711048945A CN107730499A CN 107730499 A CN107730499 A CN 107730499A CN 201711048945 A CN201711048945 A CN 201711048945A CN 107730499 A CN107730499 A CN 107730499A
- Authority
- CN
- China
- Prior art keywords
- image
- leucocyte
- nucleus
- svms
- topography
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
-
- 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
- G06V10/443—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 by matching or filtering
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (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)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of leucocyte classification method based on nu SVMs, Color Blood micro-image is pre-processed using median filter method first, then by its starting color space reflection to HLS color spaces, a width tone images after being converted, then the gray-scale image segmentation method based on nu SVMs is used to carry out coarse segmentation to the tone images, all leucocytes are detected using screening strategy layer by layer and Mathematical Morphology Method, each Leukocyte Image is finely divided again and cut, complete nucleus, the separation of cytoplasm and background, to each leucocyte, nucleus and cytoplasm extract most representational 47 features, finally the classification to leucocyte is completed by nu SVMs.The present invention can significantly improve the performance of whole leucocyte automatic identification number system.
Description
Technical field
The present invention relates to a kind of leucocyte classification method based on nu- SVMs, belong to Medical Image Processing
Field.
Background technology
Tested by the change to all kinds of quantity of leucocyte and form in blood, can usually be provided for diagnosis
Valuable information, help to make a definite diagnosis some diseases.The new doctor such as quantitative cytology, molecular biology, cellular immunology
The appearance of credit branch so that the requirement that quickly and accurately Quantitative Study is carried out to cell seems more urgent.However, by
Expert is with the naked eye examined by microscope, is wasted time and energy, and workload is very heavy, and experience, fatigue of the identification error by expert
The subjective factors such as degree have a great influence.With developing rapidly for computer image processing technology, pattern-recognition and neutral net, profit
Aided in carrying out blood cell shape identification with these advanced technologies and counted the certainty for having become blood test technology and developing
Trend.Research shows both at home and abroad, Leukocyte Image segmentation, i.e., by nucleus, cytoplasm and background separation, be whole leucocyte from
A most basic and the most key link, its Stability and veracity directly influence the identification of system in dynamic identifying system
Accuracy rate and the speed of service.Reason is that the objective factors such as illumination, dyeing can cause the image quality of cell microscopic image to decline,
And it is difficult to control.So same leucocyte is possible under different external conditions in terms of color, background, or even particle
Performance is different.Sometimes lack of standardization due to operating, the leucocyte in micro-image may be polluted by spot.In addition illumination, dyeing
Inconsistency etc. factor causes mutual difference to become more difficult.
The content of the invention
Goal of the invention：For problems of the prior art, the present invention provides a kind of based on the white of nu- SVMs
Cell sorting method.Methods described significantly improves the performance of whole leucocyte automatic identification number system, mitigates doctor significantly
The labor intensity of raw diagosis, improves diagnostic accuracy, is easy to carry out fast and accurately Quantitative Study to cell.
Technical scheme：In order to solve the above-mentioned technical problem, the technical solution adopted in the present invention is：
A kind of leucocyte classification method based on nu- SVMs, comprises the following steps：
Step A, gather Color Blood micro-image data；
Step B, the micro-image data obtained to step A carry out medium filtering, obtain medium filtering image.
Step C, the medium filtering image that step B is obtained is mapped to HLS color spaces, obtains tone images；HLS
(Hue, Lightness, Saturation tone, brightness, saturation degree) model is a kind of conventional visual color model.
Step D, the tone images obtained to step C are entered using the gray-scale image segmentation method based on nu- SVMs
Row segmentation, obtains coarse segmentation image；
Step E, the coarse segmentation image obtained to step D, use Fuzzy Cellular Neural Networks (Fuzzy Cellular
Neural Network-FCNN) detection leucocyte area image therein；
Step F, each leucocyte area image obtained to step E, using clustering methodology threshold value, with reference to threshold value
Segmentation and binary morphology method, which are finely divided, cuts, and obtains nucleus topography, cytoplasm topography and background image；
Step G, the nucleus topography obtained to step F and cytoplasm topography extract most representational 47
Individual feature；
Step H, using 47 features that step G is obtained as input vector, completed using nu- SVMs to leucocyte
Identification and classification；
Step I, treat that whole leucocyte area images that step E is obtained are disposed, count and export what step A was obtained
The final classification result of view data.
In step D, the process of the tone images coarse segmentation is as follows：
Step D-1, the tone images obtained to the step C build a histogram；
Step D-2, the histogram obtained to step D-1 by nu- SVMs carry out Function Fitting, find support to
Quantity set；
Step D-3, adaptively selected threshold value is concentrated in the supporting vector that step D-2 is found, i.e., according to the one of matched curve
Order derivative information, selection is positioned at negative value to the supporting vector near transition flex point as threshold value；
Step D-4, row threshold division is entered to the obtained tone images of step C with the step D-3 threshold values obtained.
In step G, the process of the nucleus topography and the extraction of cytoplasm local image characteristics is as follows：
Step G-1, the nucleus topography obtained to step F and cytoplasm topography extract 7 morphological feature ginsengs
Number, with quantitative description leucocyte, the number of sheets of nucleus, shape, size, the regular degree of profile；
Step G-2, the nucleus topography obtained to step F and cytoplasm topography extract 24 color property ginsengs
Number, with the brightness of quantitative description leucocyte, nucleus and cytoplasm, tone, saturation degree；
Step G-3, the nucleus topography obtained to step F extracts 16 statistic texture parameters, thin with quantitative description
The textural characteristics of karyon.
Beneficial effect：Compared with prior art, the leucocyte automatic identification provided by the invention based on nu- SVMs
Method, it is complete by the gray-scale image segmentation method based on nu- SVMs using the hue information of blood microscopic image feature
Into the coarse segmentation of tone images；All leucocytes are detected by FCNN；Using clustering methodology threshold value, with reference to Threshold segmentation
The topography comprising mononuclear leukocyte is finely divided respectively with binary morphology method and cut；In the Local map that back obtains
As on the basis of, most representational leucocyte feature is extracted, including the class of form, colour and texture etc. three totally 47 features；
The identification and classification to leucocyte are completed using nu- SVMs.This method Classification and Identification effect is preferable, and stability is high, tool
There is preferable robustness.Valuable information is provided for diagnosis, helps to carry out quickly and accurately quantitative analysis to cell
Research.
Brief description of the drawings
Fig. 1 is the flow chart of the leucocyte classification method based on nu- SVMs of the present invention；
Fig. 2 is that nucleus concavity parameter calculates schematic diagram.
Embodiment
With reference to specific embodiment, the present invention is furture elucidated, it should be understood that these embodiments are merely to illustrate the present invention
Rather than limitation the scope of the present invention, after the present invention has been read, various equivalences of the those skilled in the art to the present invention
The modification of form falls within the application appended claims limited range.
As shown in figure 1, the leucocyte classification method based on nu- SVMs of the present invention, its step are as follows：
Step 101, Color Blood micro-image data are gathered；
Step 102, the micro- input image data obtained to step 101 pre-processes；
Step 103, the pretreated image obtained to step 102 carries out HLS color space conversions, obtains tone illustration
Picture；
Step 104, the tone images obtained to step 103, with the gray level image segmentation side based on nu- SVMs
Method splits leucocyte, obtains coarse segmentation image；
Step 105, the coarse segmentation image obtained to step 104, all leucocyte area images are detected using FCNN；
Step 106, each leucocyte area image obtained to step 105, using clustering methodology threshold value, with reference to
Threshold segmentation and binary morphology method are finely divided and cut, and obtain nucleus topography, cytoplasm topography and Background
Picture；
Step 107, the nucleus topography obtained to step 106 and cytoplasm topography extract most representative
47 features；
Step 108, using 47 features that step 107 obtains as input vector, dialogue is completed using nu- SVMs
The identification and classification of cell；
Step 109, treat that all leucocyte topographies are disposed and export statistical result.
The following detailed description of the leucocyte classification method based on nu- SVMs of the present invention.
1. input micro-image
Input a width Color Blood micro-imageN is the number of pixel in image, I (xi) it is pixel xi's
Pixel vectors.
2. pretreatment
Here medium filtering is used.
3. color space is changed
Pass through substantial amounts of comparative experiments, it has been found that change of the chrominance component to illumination in HLS is insensitive, to difference
The cell microscopic image that the coloring agent of color obtains can keep good uniformity, contribute to subsequent treatment.So we will
Input picture is mapped to HLS color spaces by RGB color space.The conversion method of wherein chrominance component is as follows：
If each components range of the color value (r, g, b) of rgb space is [0,1 ..., 255], order
V '=max (r, g, b), u '=min (r, g, b) and
H=60h '
H is chrominance component value.
Tone images updating formula is：
Wherein, H represents the tone value after correction, corrected value htFor the respective value of peak value in the histogram of tone images.
Grey linear transformation, enhanced contrast effect are carried out to tone images.
4. coarse segmentation
The histogram of tone images is firstly generated, the entry deletion for being zero by wherein value, remaining nonzero term composition is finally
Histogram.
Regard histogram as a functional relation, sparse supporting vector collection is found by nu- SVMs.For
Nu- support vector regressions, due to each component w in weight vector wiAll correspond to an adjustable regularization parameter λi, this causes us
Sparse solution can be obtained, i.e. most of component of w is 0.And we claim non-zero component wiCorresponding tone value hiFor supporting vector.
Supporting vector collection has 2 excellent characteristics.First, it can depict the characteristic of original histogram well.Supporting vector is past
Toward near local maxima smallest point, for image segmentation, several suitably chosen in them are enough as threshold value
Meet that segmentation requires.Second, supporting vector collection is generally very sparse, only accounts for whole sample number n sub-fraction.Supporting vector
It is several this openness to allow us to choose segmentation threshold only from a small amount of supporting vector to ensure the efficiency of segmentation.
When the element number that supporting vector is concentrated is more than it is expected, also need further to screen.Specifically, it is exactly basis
The first derivative information of matched curve, selection is positioned at negative value to the supporting vector near transition flex point as threshold value.With institute
Obtained threshold value makees Threshold segmentation to tone images.
5. leucocyte detects
In blood microscopic image, in addition to leucocyte, also in the presence of some secondary image-regions, as red blood cell, blood coagulation are thin
Born of the same parents and spot etc..They have significant difference in color and shape with leucocyte.The gray value in leucocyte region is typically thinner than red
Born of the same parents region it is small, and because leucocyte's nuclear edge is embedded in in connective cytoplasm area, leucocyte is in bulk；Red blood cell passes through
Coarse segmentation link, typically only surplus marginal portion, annularly.So this step seeks to exclude these interference, it is complete to extract edge
Whole leucocyte.
FCNN is the useful tool for solving image processing problem.FCNN can accomplish to take into account tone during same processing
Information and structure knowledge.This is it is contemplated that using one of the reason for FCNN.Another major reason is exactly that FCNN schemes in real time
As there is unique advantage in terms of processing, it is easy to hardware realization, this can undoubtedly have very great help to improving system processing speed.This
In with FCNN realize Morphological Grayscale Reconstruction, it is as follows using parameterized template：
B=0, Afmin=without definition, Afmax=without definition, Bfmin=without definition, Bfmax-=
0,RxImage after=1, I=0, u=coarse segmentation, x0=any, y=tone images；
FCNN effects, disposably eliminate red blood cell remaining in image, spot and blood platelet etc. as then passing through
Interference, effect is fine.In the case of similar in some red blood cells region and leucocyte cytoplasmic domains gray value, also still
Leucocyte area image can be obtained well.
The first moment of gray value such as calculate each leucocyte region respectively, make each cell coordinate position, and using it in
The heart, window is set according to leucocyte maximum radial size adaptation, recovers tone images in window, so can once extract and regard
Multiple mononuclear leukocyte area images of Yezhong.
6. area image subdivision is cut
To mononuclear leukocyte area image, clustering methodology threshold value Tn and Tc is used to realize three-valued segmentation, Tn
The threshold value between nucleus and cytoplasm is represented, Tc represents the threshold value between cytoplasm and background.The advantages of this method, is, takes
Population variance is criterion in class, and it is constantly present minimum value in Tn and Tc spans, i.e., can provide optimal threshold.
So that population variance reaches minimum in class Tn and Tc are optimal threshold.The now gray scale between nucleus, cytoplasm and background
The inter-class variance of value reaches maximum.
R corrosion is carried out to the bianry image after removal background, then can so be obtained for d expansion of young shoot progress with remaining
To the bianry image for removing bulk noise jamming, general r is less than 5, d and is less than 5；Using form factor (area and the circle in region
Degree) constitutive characteristic function carries out differentiation exclusion, to detect nuclear area.Remaining is cytoplasm region.Finally give cell
Core topography, cytoplasm topography and background image.
7. feature extraction
Feature extraction is the quantitative description to cell, occupies very important status in the automatic assorting process of cell,
Directly influence the discrimination of categorizing system.Following two category feature is typically can extract to be identified：Mathematical modeling feature and structure
Feature.Method for extracting characteristics of image with mathematical modeling, the key of Classification and Identification is the extraction and selection of feature.Feature is selected
Whether appropriately to select, the effect of Classification and Identification will be directly influenced.For Leukocyte Image, the feature that can be extracted is a lot, simultaneously
Mode is also versatile and flexible.Key is to find the maximally effective invariant features parameter using the separability of classification as criterion.Namely
Say, those most representative attributes should be selected as feature.Under the guidance of clinical cytology pathology expert, with reference to cytological map
On the basis of composing and observing a large amount of actual cell images, be selectively extracted from numerous features 47 it is most representative
Parameter, establish the mathematical modelings of individual features so that computer carries out quantitative analysis.
(1) morphological feature parameter
They be to cell, the number of sheets of nucleus, shape, size, the regular degree of profile quantitative description.
(1a) cell area G1Pixel sum in=cell topography.
(1b) cytoplasm compares G with cell area2Pixel in sum of all pixels/cell topography in=cytoplasm topography
Sum.To lymphocyte G2It is smaller, and it is then larger to monocyte.
(1c) cell circularity G3=cell outline pixel count square/(in 4 π × cell topography pixel sum).
It is the key character parameter that lymphocyte distinguishes over other a few class cells, to the lymphocyte value close to 1；And neutral band form nucleus
Granulocyte, monocyte are then minimum.
The core number of sheets G of (1d) cell4The number of=karyolobism.It is thin that this is that neutrophilic segmented granulocyte distinguishes over other several classes
The important characteristic parameter of born of the same parents.Centering segmented granulocyte G4Between 2~5；Band form neutrophilic granulocyte, monokaryon and lymph
Cell not leaflet G4For 1；And eosinophil and basophilic granulocyte G4Less than 3.
(1e) its nucleolus degree G5=nucleus wire-frame image prime number square/(pixel in 4 π × nucleus topography
Sum), the same G of meaning3.Lymphocyte G5Close to 1；It is and then minimum to neutrophilia stab cell, monocyte.
The elongation of (1f) nucleus.To describe the strip of neutrophilia rhabdocyte core, core elongation degree of coming is defined
Amount.
G6=Dmax/Dmin
Wherein Dmax、DminMaximum, the minimum value that nucleus topography projects in all directions are represented respectively.This is
Distinguish band form neutrophilic granulocyte, lymphocyte, the key character of monocyte, centering stab cell G6For maximum.
(1g) nucleus concavity.Because the nucleus of monocyte is kidney-shaped, it is therefore necessary to provide the degree of concavity
Amount method.G7=1- ρimax(θ1,θ2), wherein ρi=1/180 °, illustrate that algorithm is as follows with reference to Fig. 2：Nucleus part is found out first
The symmetry axis AB of image.If symmetry axis is not present, near symmetrical axle is made with the axle that symmetric difference is minimum.Then C, 2 points of D are found out,
Make their tangent line vertical with the tangent line of A points, as C is not unique, then take intermediate value.Then point G, H are made, makes their tangent line and A
The tangent line of point is parallel.Next F, E are made, madeI, J are made, is madeFinally
Obtain the angle theta of the tangent line of E points and the tangent line of F points1, then obtain the angle theta of the tangent line of I points and the tangent line of J points2。
This category feature is more directly perceived, is easy to find and extracts.The typical leucocyte larger for distinguishing morphological differences, than
Such as lobulated granulocyte, rhabdocyte, lymphocyte best results, and then seem helpless to distinguishing granular cell, effect
It is poor.So other kinds of feature must also be extracted.
(2) color property parameter
The brightness of different type leucocyte is different, and this is reflected in pattern corresponding on the histogram of cell luminance picture not
Together, such as gray scale is inclined to, peak valley number is how many, peak value size.Tone also has similar feature with saturation degree.Therefore, colour can be used
Characteristic parameter describes its characteristic.We extract from the histogram of cell luminance picture, tone images and saturation degree image respectively
Following 8 kinds of parameters, altogether 24 color properties：Cytoplasm average value；Cytoplasm variance；Nucleus average value；Nucleus variance；
The average value of cell；The variance of cell；Caryoplasm integrates ratio；The ratio between excursion of cell and nucleus.
(3) textural characteristics parameter
Textural characteristics play an important role because of the important information arranged comprising cell tissue surface texture in identification.With
Other category features are compared, and it can preferably reflect the macroscopic view and micro-structural properties of cell image.It is adapted to below for three kinds white thin
The method of born of the same parents' analyzing image texture, we are extracted 16 statistic texture parameters from three transformation matrixs.They are from cell
Extracted in core topography.These three image transformation matrixs are defined as follows：
(3a) gray variance Correlation Matrix：Matrix element is defined as the δ neighborhood local variance u of certain picture point in image and in θ side
The probability that distance occurs jointly in the picture for the δ neighborhood local variances v of d picture point upwards.This gust of advantage is to overcome feature
The shortcomings that to greyscale-sensitive, it is not influenceed by the cell dyeing depth and image input illumination condition, only with the part side of image
Difference correlation, it is unrelated with its gray scale absolute value.Local variance reflects local gray level rate of change, and local gray level is represented not as variance is big
Uniformly, texture is thin；On the contrary, variance is small, illustrate it is open grain.Have less and larger blue-black coloured particles in basophilla granulocyte,
Also often cover karyon and be in open grain；To monokaryon, lymphocyte, area grayscale is more uniform, shows as close grain；Acidophilia core
Fallen between in granulocyte endochylema full of transparent intensive little particle.To reflect the difference on these textures, from normalization
5 angle second moment, contrast square, entropy, contrast and coefficient correlation features are extracted in matrix afterwards.In order to extract invariable rotary
Amount, we take 0 °, 45 °, 90 °, and the average of the characteristic value of 135 ° of four directions represents this 5 textural characteristics.
(3b) gray variance gradient Correlation Matrix：Matrix element is defined as in normalized gray variance image and normalized
In gradient image, picture point logarithm that some gray variance value and some Grad occur jointly.Gradient image therein is to use
Gradient operator is acted on Normalized Grey Level variance image and obtained.Gray variance gradient Correlation Matrix feature is that it has embodied a concentrated reflection of figure
Picture gray scale and image structure information, but it is unrelated with its gray scale absolute value.For coarse grained image, such as basophilla granulocyte figure
The larger particle as in, the element in matrix are distributed close to gray scale axle, and for close grain, as monocyte and neutral karyosome are thin
Born of the same parents' image, then leave gray scale axle and scattered distribution along gradient direction of principal axis.We are extracted big (small) ladder from the matrix after normalization
7 kinds of degree advantage, gray scale (gradient) nonunf ormity, entropy and contrast textural characteristics.
(3c) neighbour gray scale Correlation Matrix：The feature and the Space Rotating of image extracted from this gust and the linear change of gray value
Change unrelated, this haves a great attraction in the actual identification of cell.Element definition is in matrix：Gray scale is k, distance in image
In all neighborhood pixels less than d, the probability of pixel appearance of the gray value difference no more than a.Carried from the matrix after normalization
4 large and small several weighted volumes, the numerical value uniformity and second moment invariable rotary measure features are taken.
Textural characteristics reflect the particle properties in nucleus, size, distribution density and the nuclear staining structure of such as particle,
The differentiation of acidophilus, basophilic and neutral class granular cell in leucocyte relies primarily on these features.
8. classification and identification
Carry out quantitative analysis using nu- SVMs, 47 dimensional feature vectors that back is obtained as input vector,
Type judgement is made to leucocyte to be identified.
9. statistical result simultaneously exports
Count all kinds of leucocytes percentage in blood microscopic image, display or printing analyze data result.
Pass through above-mentioned embodiment, it is seen that the invention has the advantages that：
(1) this method completes the coarse segmentation of tone images using the gray-scale image segmentation method based on nu- SVMs,
Mainly by introducing nu- SVMs, limited sparse supporting vector collection is obtained while fitting, then directly therefrom
Filter out required segmentation threshold.This method is split suitable for photo chromic microimage, can effectively overcome illumination, dyeing etc. objective
The interference of factor, have the advantages that segmentation effect is excellent, computational efficiency is high, parameter setting is easy, be advantageous to subsequent characteristics extract with
Differential counting, solid foundation is established to improve the recognition accuracy of whole system.
(2) according to the experience of clinical cytology scholar, present invention extraction more than human eye can differentiate more than three classes totally 47 in vain
Cell characteristic parameter, and use nu- SVMs to realize the automatic classification of six class leucocytes, classifying quality is preferable, stability
Height, there is preferable robustness.
Claims (6)
1. a kind of leucocyte classification method based on nu- SVMs, it is characterised in that comprise the following steps：
Step A, gather Color Blood micro-image data；
Step B, the micro-image data obtained to step A carry out medium filtering, obtain medium filtering image.
Step C, the medium filtering image that step B is obtained is mapped to HLS color spaces, obtains tone images；
Step D, the tone images obtained to step C are divided using the gray-scale image segmentation method based on nu- SVMs
Cut, obtain coarse segmentation image；
Step E, the coarse segmentation image obtained to step D, use Fuzzy Cellular Neural Networks (Fuzzy Cellular Neural
Network-FCNN) detection leucocyte area image therein；
Step F, each leucocyte area image obtained to step E, using clustering methodology threshold value, with reference to Threshold segmentation
It is finely divided and cuts with binary morphology method, obtains nucleus topography, cytoplasm topography and background image；
Step G, the nucleus topography obtained to step F and cytoplasm topography extract most representational 47 spies
Sign；
Step H, using 47 features that step G is obtained as input vector, the knowledge to leucocyte is completed using nu- SVMs
Not with classification；
Step I, treat that whole leucocyte area images that step E is obtained are disposed, count and export the image obtained to step A
The final classification result of data.
A kind of 2. leucocyte classification method based on nu- SVMs according to claim 1, it is characterised in that step
In rapid D, the process of the tone images coarse segmentation is as follows：
Step D-1, the tone images obtained to the step C build a histogram；
Step D-2, the histogram obtained by nu- SVMs to step D-1 carry out Function Fitting, find supporting vector
Collection；
Step D-3, adaptively selected threshold value is concentrated in the supporting vector that step D-2 is found, i.e., is led according to the single order of matched curve
Number information, selection is positioned at negative value to the supporting vector near transition flex point as threshold value；
Step D-4, row threshold division is entered to the obtained tone images of step C with the step D-3 threshold values obtained.
A kind of 3. leucocyte classification method based on nu- SVMs according to claim 1, it is characterised in that step
In rapid G, the process of the nucleus topography and the extraction of cytoplasm local image characteristics is as follows：
Step G-1, the nucleus topography obtained to step F and cytoplasm topography extract 7 morphological feature parameters, with
Quantitative description leucocyte, the number of sheets of nucleus, shape, size, the regular degree of profile；
Step G-2, the nucleus topography obtained to step F and cytoplasm topography extract 24 color property parameters,
With the brightness of quantitative description leucocyte, nucleus and cytoplasm, tone, saturation degree；
Step G-3, the nucleus topography obtained to step F extracts 16 statistic texture parameters, with quantitative description nucleus
Textural characteristics.
4. a kind of leucocyte classification method based on nu- SVMs according to claim 2, it is characterised in that raw
Into the histogram of tone images, the entry deletion for being zero by wherein value, remaining nonzero term forms final histogram；
Regard histogram as a functional relation, sparse supporting vector collection is found by nu- SVMs；Work as supporting vector
When the element number of concentration is more than it is expected, also need further to screen, according to the first derivative information of matched curve, selection is located at
Negative value is to the supporting vector near transition flex point as threshold value.
A kind of 5. leucocyte classification method based on nu- SVMs according to claim 3, it is characterised in that shape
State characteristic parameter include cell area, cytoplasm and cell area than, the core number of sheets, its nucleolus of cell circularity, cell
Degree, the elongation of nucleus and nucleus concavity.
A kind of 6. leucocyte classification method based on nu- SVMs according to claim 3, it is characterised in that from
16 statistic texture parameters are extracted in three transformation matrixs, three image transformation matrixs are defined as follows：
(3a) gray variance Correlation Matrix：Matrix element is defined as the δ neighborhood local variance u of certain picture point in image and on θ direction
The probability that distance occurs jointly in the picture for the δ neighborhood local variances v of d picture point；
(3b) gray variance gradient Correlation Matrix：Matrix element is defined as in normalized gray variance image and normalized gradient
In image, picture point logarithm that some gray variance value and some Grad occur jointly；
(3c) neighbour gray scale Correlation Matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711048945.0A CN107730499A (en) | 2017-10-31 | 2017-10-31 | A kind of leucocyte classification method based on nu SVMs |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711048945.0A CN107730499A (en) | 2017-10-31 | 2017-10-31 | A kind of leucocyte classification method based on nu SVMs |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107730499A true CN107730499A (en) | 2018-02-23 |
Family
ID=61202928
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711048945.0A Pending CN107730499A (en) | 2017-10-31 | 2017-10-31 | A kind of leucocyte classification method based on nu SVMs |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107730499A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805865A (en) * | 2018-05-22 | 2018-11-13 | 杭州智微信息科技有限公司 | A kind of myeloplast localization method based on saturation degree cluster |
| CN109087276A (en) * | 2018-05-17 | 2018-12-25 | 苏州斯玛维科技有限公司 | The component of radioscopic image based on SMT charging tray counts automatically and localization method |
| CN110135271A (en) * | 2019-04-19 | 2019-08-16 | 上海依智医疗技术有限公司 | A kind of cell sorting method and device |
| CN110211108A (en) * | 2019-05-29 | 2019-09-06 | 武汉兰丁医学高科技有限公司 | A kind of novel abnormal cervical cells automatic identifying method based on Feulgen colouring method |
| CN111458269A (en) * | 2020-05-07 | 2020-07-28 | 厦门汉舒捷医疗科技有限公司 | Artificial intelligent identification method for peripheral blood lymph micronucleus cell image |
| CN112183210A (en) * | 2020-08-28 | 2021-01-05 | 江苏康尚生物医疗科技有限公司 | Sparse matrix-based particle automatic focusing method and system and storage medium |
| CN113723505A (en) * | 2021-08-30 | 2021-11-30 | 上海大学 | Cell image classification method based on cytoskeleton image quantitative analysis and classifier thereof |
| CN116758068A (en) * | 2023-08-17 | 2023-09-15 | 天津市肿瘤医院（天津医科大学肿瘤医院） | Marrow picture cell morphology analysis method based on artificial intelligence |
| CN118314570A (en) * | 2024-06-11 | 2024-07-09 | 重庆医科大学绍兴柯桥医学检验技术研究中心 | White blood cell classification method and system for error correction of detection card |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020639A (en) * | 2012-11-27 | 2013-04-03 | 河海大学 | Method for automatically identifying and counting white blood cells |
| CN103679184A (en) * | 2013-12-06 | 2014-03-26 | 河海大学 | Method for leukocyte automatic identification based on relevant vector machine |
| CN103745210A (en) * | 2014-01-28 | 2014-04-23 | 爱威科技股份有限公司 | Method and device for classifying white blood cells |
-
2017
- 2017-10-31 CN CN201711048945.0A patent/CN107730499A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020639A (en) * | 2012-11-27 | 2013-04-03 | 河海大学 | Method for automatically identifying and counting white blood cells |
| CN103679184A (en) * | 2013-12-06 | 2014-03-26 | 河海大学 | Method for leukocyte automatic identification based on relevant vector machine |
| CN103745210A (en) * | 2014-01-28 | 2014-04-23 | 爱威科技股份有限公司 | Method and device for classifying white blood cells |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109087276A (en) * | 2018-05-17 | 2018-12-25 | 苏州斯玛维科技有限公司 | The component of radioscopic image based on SMT charging tray counts automatically and localization method |
| CN109087276B (en) * | 2018-05-17 | 2022-02-01 | 苏州斯玛维科技有限公司 | Automatic counting and positioning method for components based on X-ray image of SMT tray |
| CN108805865A (en) * | 2018-05-22 | 2018-11-13 | 杭州智微信息科技有限公司 | A kind of myeloplast localization method based on saturation degree cluster |
| CN110135271A (en) * | 2019-04-19 | 2019-08-16 | 上海依智医疗技术有限公司 | A kind of cell sorting method and device |
| CN110211108A (en) * | 2019-05-29 | 2019-09-06 | 武汉兰丁医学高科技有限公司 | A kind of novel abnormal cervical cells automatic identifying method based on Feulgen colouring method |
| CN111458269A (en) * | 2020-05-07 | 2020-07-28 | 厦门汉舒捷医疗科技有限公司 | Artificial intelligent identification method for peripheral blood lymph micronucleus cell image |
| CN112183210A (en) * | 2020-08-28 | 2021-01-05 | 江苏康尚生物医疗科技有限公司 | Sparse matrix-based particle automatic focusing method and system and storage medium |
| CN113723505A (en) * | 2021-08-30 | 2021-11-30 | 上海大学 | Cell image classification method based on cytoskeleton image quantitative analysis and classifier thereof |
| CN113723505B (en) * | 2021-08-30 | 2024-04-23 | 上海大学 | Cell image classification method based on cytoskeletal image quantitative analysis and classifier thereof |
| CN116758068A (en) * | 2023-08-17 | 2023-09-15 | 天津市肿瘤医院（天津医科大学肿瘤医院） | Marrow picture cell morphology analysis method based on artificial intelligence |
| CN116758068B (en) * | 2023-08-17 | 2023-11-07 | 天津市肿瘤医院（天津医科大学肿瘤医院） | Marrow picture cell morphology analysis method based on artificial intelligence |
| CN118314570A (en) * | 2024-06-11 | 2024-07-09 | 重庆医科大学绍兴柯桥医学检验技术研究中心 | White blood cell classification method and system for error correction of detection card |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107730499A (en) | A kind of leucocyte classification method based on nu SVMs | |
| US10565479B1 (en) | Identifying and excluding blurred areas of images of stained tissue to improve cancer scoring | |
| US10474874B2 (en) | Applying pixelwise descriptors to a target image that are generated by segmenting objects in other images | |
| Jagadev et al. | Detection of leukemia and its types using image processing and machine learning | |
| CN103020639A (en) | Method for automatically identifying and counting white blood cells | |
| CN108376400B (en) | Automatic bone marrow cell classification method | |
| CN103679184A (en) | Method for leukocyte automatic identification based on relevant vector machine | |
| US5828776A (en) | Apparatus for identification and integration of multiple cell patterns | |
| CN102682305B (en) | Automatic screening system and automatic screening method using thin-prep cytology test | |
| Ko et al. | Automatic white blood cell segmentation using stepwise merging rules and gradient vector flow snake | |
| US5987158A (en) | Apparatus for automated identification of thick cell groupings on a biological specimen | |
| CN109190567A (en) | Abnormal cervical cells automatic testing method based on depth convolutional neural networks | |
| Madhukar et al. | New decision support tool for acute lymphoblastic leukemia classification | |
| CN109635846A (en) | A kind of multiclass medical image judgment method and system | |
| Shahin et al. | A novel white blood cells segmentation algorithm based on adaptive neutrosophic similarity score | |
| CN106529592A (en) | License plate recognition method based on mixed feature and gray projection | |
| Di Ruberto et al. | A leucocytes count system from blood smear images: Segmentation and counting of white blood cells based on learning by sampling | |
| CN108710916A (en) | The method and device of picture classification | |
| Fang et al. | On the accurate counting of tumor cells | |
| Sobrevilla et al. | White blood cell detection in bone marrow images | |
| Apou et al. | Detection of lobular structures in normal breast tissue | |
| CN115375690A (en) | Tongue picture putrefaction classification and identification method | |
| Ruberto et al. | A leucocytes count system from blood smear images. | |
| Lakshmi et al. | Multiple feature extraction from cervical cytology images by Gaussian mixture model | |
| Kim et al. | Nucleus segmentation and recognition of uterine cervical pap-smears |
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: 20180223 |