CN108185984A - The method that eyeground color picture carries out eyeground lesion identification - Google Patents
The method that eyeground color picture carries out eyeground lesion identification Download PDFInfo
- Publication number
- CN108185984A CN108185984A CN201711456141.4A CN201711456141A CN108185984A CN 108185984 A CN108185984 A CN 108185984A CN 201711456141 A CN201711456141 A CN 201711456141A CN 108185984 A CN108185984 A CN 108185984A
- Authority
- CN
- China
- Prior art keywords
- picture
- image
- eyeground
- blood vessel
- carries out
- 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
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000003902 lesion Effects 0.000 title claims abstract description 18
- 230000000740 bleeding effect Effects 0.000 claims abstract description 10
- 208000009857 Microaneurysm Diseases 0.000 claims abstract description 8
- 230000000694 effects Effects 0.000 claims abstract description 8
- 210000004204 blood vessel Anatomy 0.000 claims description 39
- 238000012549 training Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 11
- 210000003733 optic disk Anatomy 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 9
- 230000000877 morphologic effect Effects 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000009826 distribution Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 238000010801 machine learning Methods 0.000 claims description 4
- 238000007637 random forest analysis Methods 0.000 claims description 4
- 241001156002 Anthonomus pomorum Species 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 206010002329 Aneurysm Diseases 0.000 claims description 2
- 239000008280 blood Substances 0.000 claims description 2
- 210000004369 blood Anatomy 0.000 claims description 2
- 238000005520 cutting process Methods 0.000 claims description 2
- 238000003708 edge detection Methods 0.000 claims description 2
- 238000005530 etching Methods 0.000 claims description 2
- 230000005484 gravity Effects 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000003860 storage Methods 0.000 claims description 2
- 230000002792 vascular Effects 0.000 claims 2
- 239000003973 paint Substances 0.000 claims 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 9
- 201000010099 disease Diseases 0.000 abstract description 8
- 238000003745 diagnosis Methods 0.000 abstract description 6
- 206010012689 Diabetic retinopathy Diseases 0.000 abstract description 3
- 238000005259 measurement Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 5
- 239000002956 ash Substances 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 206010012601 diabetes mellitus Diseases 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 235000002918 Fraxinus excelsior Nutrition 0.000 description 1
- 208000010412 Glaucoma Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000686 essence Substances 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 235000021384 green leafy vegetables Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000000059 patterning Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/0016—Operational features thereof
- A61B3/0033—Operational features thereof characterised by user input arrangements
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/0016—Operational features thereof
- A61B3/0041—Operational features thereof characterised by display arrangements
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
- A61B3/1241—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes specially adapted for observation of ocular blood flow, e.g. by fluorescein angiography
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/14—Arrangements specially adapted for eye photography
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color 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
- G06T2207/30041—Eye; Retina; Ophthalmic
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Ophthalmology & Optometry (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- General Physics & Mathematics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Theoretical Computer Science (AREA)
- Vascular Medicine (AREA)
- Hematology (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of methods that eyeground color picture carries out eyeground lesion identification.Automatic positioning and measurement to each position in the color picture of eyeground, achieve the effect that disease prescreening, the picture Automatic sieve for having lesion suspicion is selected, can fixation and recognition fast and effeciently be carried out to bleeding, exudation and microaneurysm, so as to diseases such as auxiliary diagnosis diabetic retinopathy, working doctor amount is reduced;And its result is independent of doctors experience, more objective, can effectively assist a physician and carry out the diagnosis of disease, realizes the purpose of remote medical consultation with specialists.
Description
Technical field
The present invention relates to a kind of methods that eyeground color picture carries out eyeground lesion identification.
Background technology
It, can be quick in order to carry out the identification of the common lesion in eyeground in the color picture of eyeground and boundary mapping in large quantity rapidly
Fixation and recognition effectively is carried out to bleeding, exudation and microaneurysm, so as to diseases such as auxiliary diagnosis diabetic retinopathy
Disease.In clinic, since the manpowers such as outlying mountain area, basic hospital oculist diagosis personnel related to eyeground are limited, such as to big
The eyeground color picture of amount is mechanically reviewed one by one, and action is heavy, single repetition and inefficient, and waste is big
The valuable human resources of amount.Existing eyeground color picture automatic recognition system also relates to eye fundus image automatic identification square partition
Method, but be not to carry out being accurately positioned for eyeground locations of structures, lesion identification.In addition existing, method is mostly to utilize online mark
Quasi- picture is carried out with reference to comparing identification, however the eyeground color picture picture in real-life clinical is not normal pictures or even including very
The all problematic picture of multi-focus, light and shade, many systems are all that directly picture is identified at present, without component tablet quality,
Because online java standard library is all the preferable picture of quality, and quantity is all seldom.
Invention content
The primary and foremost purpose of the present invention is to provide a kind of method that eyeground color picture carries out eyeground lesion identification.To in the color picture of eyeground
The automatic positioning and measurement at each position, achieve the effect that disease prescreening, the picture Automatic sieve for having lesion suspicion are selected, energy
It is enough that fixation and recognition fast and effeciently is carried out to bleeding, exudation and microaneurysm, so as to diseases such as auxiliary diagnosis diabetes, subtract
Few working doctor amount;And its result is independent of doctors experience, and more objective, can effectively assist a physician progress disease
The purpose of remote medical consultation with specialists is realized in diagnosis.
In order to solve the above technical problems, the technical solution adopted in the present invention is:
The method that eyeground color picture provided by the invention carries out eyeground lesion identification has the characteristics that:
1st, by the way that morphological method is combined with machine learning method, the region of positioning is carried out by morphological method
Then preliminary treatment carries out machine learning method prediction, is finally accurately positioned again by morphological method;
2nd, the eyeground color picture of a large amount of patients acquired to hospital or community carries out automatic identification, to assist doctor to a large amount of
Screening for diabetic retinopathy and medical center routine eyeground color picture picture are diagnosed, so as to auxiliary diagnosis diabetes
Etc. diseases;
3rd, using a large amount of Zhang Zhenshi eyeground color picture make model fully learn it is various in the case of image data feature so that
Judge more accurate, and have better fault-tolerance, simultaneity factor carries out eyeground color picture pretreatment and picture quality first
Differentiate, to ensure this system in the case where picture quality is irregular, can also have good recognition capability, there is universality.
Description of the drawings
The attached drawing for forming the part of the application is used to provide further understanding of the present invention, schematic reality of the invention
Example and its explanation are applied for explaining the present invention, is not constituted improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is the example images of selection of the embodiment of the present invention;
Fig. 2 is the intensity profile of image selected by the embodiment of the present invention;
Fig. 3 is vessel profile schematic diagram when blood vessel of the embodiment of the present invention identifies substantially;
Deburring treated vessel profile schematic diagram when Fig. 4 is identified for blood vessel of the embodiment of the present invention;
Vessel profile schematic diagram breakpoint is fitted schematic diagram when Fig. 5 is identified for blood vessel of the embodiment of the present invention;
Fig. 6 determines vessel borders schematic diagram when being identified for blood vessel of the embodiment of the present invention;
Fig. 7 is blood vessel recognition effect figure of the embodiment of the present invention;
Fig. 8 is the image of green track detail section when microaneurysm of the embodiment of the present invention identifies;
Fig. 9 is bleeding result design sketch of the embodiment of the present invention;
Figure 10, Figure 12 ooze out design sketch for the embodiment of the present invention;
Figure 11 is the corresponding original image of the embodiment of the present invention;
Figure 13 is the candidate point image of microaneurysm of the embodiment of the present invention.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, it is right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Embodiment
A kind of method that eyeground color picture carries out eyeground lesion identification, includes the following steps:
Picture quality detects:
First, feature extraction:
Using the skeleton of image, its textural characteristics, RGB, three figure layers, 15 features of each figure layer extraction are extracted.
Edge detection is carried out to image using canny operators first, then denoising is carried out with medium filtering, later with pre-
Processed figure calculates the number of the total pixel in the edge, overall circumference at edge, the maximum height of fringe region, maximum width, strange
Chain number of codes (number of edge discrete point), target area, rectangular degree, the elongation of number chain.
Followed by seven invariant moment features of extraction image:
The sum of horizontal and vertical directed variance,more
distributed towards horizontal and vertical axes,the values are enlarged.
The covariance value of vertical and horizontal axes when the
variance intensity of vertical axis and horizontal axis were similar.
The result emphasizing the values inclined to left/right and
upper/lower axes.
The result emphasizing the values counterbalancing to left/right
and upper/lower axes.
The extraction of values invariant against size,rotation,
and location.
According to the judgement of clarity, RGB, each figure layer extracts 5 features for judging clarity.
A) gray level entropy:
It reflects the number of average information in image.The one-dimensional entropy of image represents that the aggregation of intensity profile in image is special
The included information content of sign, enables piRepresent that gray value is the ratio shared by the pixel of i in image, then defines the unitary of gray level image
Gray level entropy is:
B) Brenner gradient functions
Brenner gradient functions are simplest Gradient functions, it only simply calculates two neighboring pixel ash
Square of difference is spent, which is defined as follows:
F (x, y) represents the gray value of image f corresponding pixel points (x, y).
C) variance function
It, can be using variance function as evaluation because the image clearly focused on has the gray difference than blurred picture bigger
Function:
Wherein,For the average gray value of entire image, the function pair noise is more sensitive, and image frame is purer, function
It is worth smaller.
D) energy gradient function
E) gradient function
3rd, 256 features of grey level histogram.
RGB is such as now converted into gray-scale map, then Gray=0.29900*R+0.58700*G+0.11400*B utilizes ash
Spend histogram extraction feature.
Grey level histogram is the function about grey level distribution, is the statistics to grey level distribution in image.Intensity histogram
Figure is by all pixels in digital picture, according to the size of gray value, counts the frequency of its appearance.0-255 gray value
Frequency extracts 256 features altogether.
4th, rgb space is converted, extracts color and textural characteristics 256.
Utilize paper《Color and texture descriptors》The rgb space of script is converted into HSV space,
Color histogram is calculated, has obtained 256 features.
So far, all extraction finishes for required whole features, then by the use of all features extracted as independent variable,
The quality of picture quality predicts picture quality as dependent variable (0 or 1).Here we using Random Forest model into
Row prediction.
First, machine learning is predicted:
Random forests algorithm:
1st, training dataset d=(X, y) is given, wherein X is the feature extracted, and (0 represents figure to 0,1 classified variables of y
Tablet quality is poor, and 1 represents that picture quality is good).Fixed m≤p (m is the Characteristic Number randomly selected out, and p is characterized total number) and
Set the number B of (decision Tree algorithms).
2nd, following steps are done to each b=1,2 ..., B:
A) bootstrap training sets are constructed by extracting n times from n sample at random to training data d
B) it usesIn data configuration depth capacity treeM are extracted from p variable at random into line splitting;
C) storage tree and the information of bootstrap samples.
3rd, to arbitrary future position x0, carry out the fitting and prediction of random forest.To each treeOne will be predicted
Classification, in this way due to there is B tree, it is possible to predict B 01 classifications.Final prediction result is exactly in B classification, goes out
The most classification of occurrence number (0 or 1).
Due to the picture there is a large amount of poor qualities in true picture, so being detected by picture quality first by this batch of matter
It measures bad picture to screen, subsequent processing is carried out just for the eye fundus image that picture quality reaches a standard.
Image preprocessing:
Pretreatment is using histogram equalization.Select a recognition effect best first in all images, such as Fig. 1
It is shown, and the intensity profile of its tri- track of RGB is extracted, as shown in Figure 2.As standard drawing.
Optic disk identifies:
Optic disk identification is broadly divided into three key steps:Just positioning (ROI extractions), is accurately positioned, Smoothing fit
Just positioning:Being primarily based on optic disk has the characteristics that highlight, and the most apparent on red track, we choose first
Red track is analyzed.Specifically red green track can visually identify optic disk position, and red track becomes apparent.It is interested
Region (ROI) extraction mainly utilizes the method for adaptive threshold fuzziness.It first will be by the brighter region threshold value of whole pictures
The method of segmentation extracts, remaining darker area is filled up using mean value.Modified picture carries out threshold value cutting again.Pass through
Successive ignition reduces brighter areas area step by step.After ROI areas pre-determined threshold value, stop iteration.Then it is right again
The highlight regions extracted are screened.Then it extracts the center of the ROI and intercepts ROI for analyzing in next step.At ROI
It puts in determining method, also has simple threshold values patterning method as currently a popular method class:Optic cup and disc
localization for Detection of glaucoma usingMatlab,Hanamant M.Havagondi,2
Mahesh S.Kumbhar.Kaiser Window positioning modes:Blood vessel inpainting based technique
for efficient localization andsegmentationof optic disc in digital fundus
Images, Biomedical Signal Processing and Control 25 (2016) 108-117 etc..It is compared to it
His method, we are advantage:Merely using red orbit information, blood vessel influences smaller.For the photo that boundary exposure is excessive.
Method therefor of the present invention can remove this some effects quickly, and ROI will not be extracted and cause to perplex.And for a part of image sheet
Height bright area is excessive (or height lesion, optic disk brightness are inadequate), these quality are not very high, the inaccurate pictures of ROI positioning
Program meeting automatic prompt quality problems, without subsequent analysis.
It is accurately positioned and Smoothing fit:
In the ROI extracted, mainly first remove noise effect using Morphological scale-space method, then image is carried out
Threshold segmentation can obtain relatively rough boundary position.Then it is (minimum external ellipse ellipse fitting to be carried out for boundary position
Circle), fit a boundary parameter equation
X=a*cos (t) * cos (θ)-b*sin (t) * sin (θ)+x0
Y=a*cos (t) * cos (θ)-b*sin (t) * sin (θ)+y0
Wherein θ is oval inclination angle, and a, b are long semi-minor axis, and t is parameter, x0, y0It is elliptical center coordinate.Finally by boundary
Equation is plotted in artwork.At present in terms of optic disk boundary alignment, mainly there are the calculations such as fixed threshold segmentation and region growing
Method.Fixed threshold segmentation stability it is worst, Boundary Recognition is not allowed, and our adaptive threshold method then according to optic disk area from
Dynamic selection optimal threshold, will not cause the apparent erroneous judgement on optic disk boundary.And algorithm of region growing then chooses initial seed point
It there are certain requirements, and be likely to result in the problem of identification region is less than normal.
Blood vessel identifies:
One, read in picture.
Two, it is the dark border removed around round eyeground picture to first processing that picture is done.
Three, picture processing
To picture processing, vessel profile substantially is obtained.First, noise is removed in pretreatment, and medium filtering, threshold value are gone later
It makes an uproar, the results are shown in Figure 3;
Then, the blood vessel obtained to pretreatment, carries out multiple etching operation, obtains the substantially distribution of blood vessel, at this moment
Due to part figure sector-meeting because the picture quality original similar to dim spot locally occurs in picture caused by the reasons such as shooting angle, light
Cause, so the picture that this step obtains is it is possible that the situation that blood vessel is broken in certain, so not connecting to what is obtained then
The blood vessel of continuous fracture carries out pixel link and diagonal filling, obtains continuous blood vessel.Then the coarse blood vessel to obtaining
Deburring processing is carried out, to the general profile of blood vessel, the results are shown in Figure 4.
Four, blood vessel identification
Then blood vessel identification is that each section of blood vessel is regarded as an individual, identifies each section of blood vessel.
Main is the center line for finding blood vessel first.Because the profile of tentatively identification blood vessel is not allowed, but pass through corrosion
Comparatively operation can find more accurately vessel centerline.This step is just separated each section of blood vessel, removes blood
The tie point of pipe, each section is all continuous line.Then breakpoint is fitted using algorithm, as shown in Figure 5.
After center line is chosen, changed using the outside shade of gray in center line both sides, determine vessel borders, as shown in Figure 6.
Finally the blood vessel of identification is drawn in original picture, as shown in Figure 7.
Bleeding exudation identification:
Bleeding is oozed out also with eyeground picture and its corresponding picture row for containing only the bleeding seepage areas marked
What training formed.
Here the grader used is Adaboost.Feature selecting during training and feature selecting during training arteriovenous
Angle is similar, and trained process and the process for obtaining grader are also identical.
One, it is also to remove the outermost black surround of picture that picture, which is carried out, first in identification.
Two, the feature of picture is read, feature herein is identical with feature during training grader.
Three, rough candidate region is first selected with morphologic method.
Here approximate region is determined by gray scale merely first, Morphological scale-space removal is then carried out to selected region and is made an uproar
Sound.
Four, it goes to classify with trained grader.It obtains as a result, wherein bleeding result is as shown in Fig. 9, Figure 12, exudation is tied
Fruit is as shown in Figure 10, Figure 12.(Figure 12 greens represent bleeding, and blue represents exudation.Its corresponding original image is Figure 11)
Microaneurysm identifies:
The green track of pending picture is extracted first, it, will be filtered then to green track using medium filtering
Image is as background, then the gray scale of background is subtracted to the gray scale of artwork, can obtain the image of prominent green track detail section,
As shown in Figure 8.
Then, using maximum mutual information method, the formula of calculating is as follows:
P in formula is the correlation matrix of 256 gray values of each image.Correlation is stronger, represents two ashes
The probability neighbouring in the picture of degree is bigger.
The adaptive threshold value for choosing Fig. 8 carries out binaryzation, reuses artwork extraction angiosomes, by angiosomes and and
Blood vessel publishes in instalments region removal together, obtains the candidate point image of all microaneurysms, as shown in figure 13.
Using all candidate points microangioma region doubtful as one, and centered on the center of gravity of candidate point, carry out
Feature extraction.Feature includes the red track mean value, standard deviation, minimax value of small picture, mean value, the standard of green track
Difference, minimax value, green track are mapped to the mean value, standard deviation, minimax value of the small picture of 0-256, the face of connected domain
Product, the minimum enclosed rectangle length and width of connected domain, the entropy of image, the gradient of image, the center of image and the four of corner regions
The information such as average gray.These features are all exported, and is labeled with being labeled as brigadier per a sample.
All data with picture are divided, the positive sample of training set is expanded using SMOTE methods.Finally use
Xgboost models are trained training set.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
All any modification, equivalent and improvement made within refreshing and principle etc., should all be included in the protection scope of the present invention.
Claims (6)
1. a kind of method that eyeground color picture carries out eyeground lesion identification, it is characterised in that include the following steps:
Picture quality detects, and inputs original image, carries out the extraction of characteristics of image, and be trained, and picture quality is carried out pre-
It surveys, detects that the eye fundus image that picture quality reaches a standard carries out subsequent processing;
Image preprocessing selects a recognition effect best, and extracts the gray scale of its tri- track of RGB in all images
Distribution, as standard drawing;
Bleeding oozing of blood identifies that it is the region in the certain distance of the center of circle as target area to select using optic disk, according to where blood vessel
Position by the region using vascular wall as boundary, is divided on the outside of blood vessel, vessel borders, three sub-regions on the inside of blood vessel;
These three subregions are extracted with the feature of shade of gray variation and two broad aspects of texture variations respectively;
Picture is inputted during identification, then identifies the blood vessel of target area, then in the outside of blood vessel, vascular wall, inside point
Indescribably take the selected feature of training;
The Machine learning classifiers obtained with training are identified, and obtain as a result, and result is illustrated in original picture.
2. the method that eyeground color picture as described in claim 1 carries out eyeground lesion identification, it is characterised in that described image feature
Extraction include the following steps:
Edge detection is carried out to image using canny operators first, then denoising is carried out with medium filtering, later with pretreatment
The figure crossed, the number of the calculating total pixel in edge, the overall circumference at edge, the maximum height of fringe region, maximum width, odd number chain
Chain number of codes, target area, rectangular degree, elongation, then extract image seven invariant moment features;
Each figure layer extracts 5 features for judging clarity:Gray level entropy, Brenner gradient functions, variance function, energy gradient
Function, gradient function;
Feature is extracted using grey level histogram;
The rgb space of script is converted into HSV space, calculates color histogram.
3. the method that eyeground color picture as described in claim 1 carries out eyeground lesion identification, it is characterised in that described to picture matter
Amount carries out prediction and includes the following steps:
1) training dataset d=(X, y) is given, wherein x is the feature extracted, and y 0,1 classified variable, 0 represents picture matter
Amount is poor, and 1 represents that picture quality is good, and fixed m≤p, m is the Characteristic Number randomly selected out, and p is characterized total number and decision
The number B set in tree algorithm;
2) to each b=1,2 ..., B do following steps:
A) bootstrap training sets are constructed by extracting n times from n sample at random to training data d
B) it usesIn data configuration depth capacity treeM are extracted from p variable at random into line splitting;
C) storage tree and the information of bootstrap samples;
3) to arbitrary future position X0, the fitting and prediction of random forest are carried out, to each treeA class will be predicted
Not, in this way due to there is B tree, it is possible to predict B 01 classifications, final prediction result is exactly in B classification, occurs
The most classification of number.
4. the method that eyeground color picture as described in claim 1 carries out eyeground lesion identification, it is characterised in that further include blood vessel knowledge
Other step:
Noise is removed in picture processing, first processing, and medium filtering, threshold denoising processing, obtain vessel profile substantially later;
Vessel profile is obtained, the vessel profile pre-processed carries out multiple etching operation, obtains the substantially distribution model of blood vessel
It encloses, pixel link then is carried out to the obtained blood vessel of discontinuous fracture and diagonal is filled, obtains continuous blood vessel, then
Deburring processing is carried out to obtained coarse blood vessel, to vessel profile;
Blood vessel identifies, finds the center line of blood vessel first, is changed using the outside shade of gray in center line both sides, determines blood vessel side
The blood vessel of identification is finally drawn in original picture by boundary.
5. the method that eyeground color picture as described in claim 1 carries out eyeground lesion identification, it is characterised in that further include optic disk knowledge
Other step:
Just positioning, first will extract the method for whole pictures brighter region Threshold segmentation, remaining darker area profit
It is filled up with mean value, modified picture carries out threshold value cutting again, and by successive ignition, brighter areas area is subtracted step by step
It is small, after area-of-interest area pre-determined threshold value, stop iteration, then the highlight regions extracted are screened, carry
It takes out the center of the area-of-interest and intercepts to analyze in next step;
It is accurately positioned and Smoothing fit, first removes noise effect using Morphological scale-space method, threshold value then is carried out for image
Then segmentation carries out ellipse fitting for boundary position, finally paints absorbing boundary equation to obtain relatively rough boundary position
System is in artwork.
6. the method that eyeground color picture as described in claim 1 carries out eyeground lesion identification, it is characterised in that further include capilary
Knurl identification step:
The green track of pending picture is extracted first, then to green track using medium filtering, by filtered image
As background, then the gray scale of background is subtracted to the gray scale of artwork, obtain the image of prominent green track detail section;
Using maximum mutual information method, the formula of calculating is as follows:
P in formula is the correlation matrix of 256 gray values of each image;
The threshold value of the adaptive image for choosing green track detail section carries out binaryzation, reuses artwork extraction angiosomes,
Angiosomes and the region published in instalments together with blood vessel are removed, obtain the candidate point image of all microaneurysms;
Using all candidate points microangioma region doubtful as one, and centered on the center of gravity of candidate point, feature is carried out
Extraction, feature include the red track mean value of small picture, standard deviation, minimax value, the mean value of green track, standard deviation,
Minimax value, green track are mapped to the mean value, standard deviation, minimax value of the small picture of 0-256, the area of connected domain,
The minimum enclosed rectangle length and width of connected domain, the entropy of image, image gradient, the center of image and the four of corner regions it is flat
The information such as equal gray scale all export these features, and be labeled with being labeled as brigadier per a sample;
All data with picture are divided, the positive sample of training set is expanded using SMOTE methods, is finally used
Xgboost models are trained training set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711456141.4A CN108185984A (en) | 2017-12-28 | 2017-12-28 | The method that eyeground color picture carries out eyeground lesion identification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711456141.4A CN108185984A (en) | 2017-12-28 | 2017-12-28 | The method that eyeground color picture carries out eyeground lesion identification |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108185984A true CN108185984A (en) | 2018-06-22 |
Family
ID=62585369
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711456141.4A Pending CN108185984A (en) | 2017-12-28 | 2017-12-28 | The method that eyeground color picture carries out eyeground lesion identification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108185984A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109472781A (en) * | 2018-10-29 | 2019-03-15 | 电子科技大学 | A kind of diabetic retinopathy detection system based on serial structure segmentation |
CN109544540A (en) * | 2018-11-28 | 2019-03-29 | 东北大学 | A kind of diabetic retina picture quality detection method based on image analysis technology |
CN109829882A (en) * | 2018-12-18 | 2019-05-31 | 苏州比格威医疗科技有限公司 | A kind of stages of DR prediction technique |
CN109846513A (en) * | 2018-12-18 | 2019-06-07 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging method, system and image measuring method, processing system and medium |
CN109948622A (en) * | 2019-03-26 | 2019-06-28 | 数坤(北京)网络科技有限公司 | A kind of neck body aneurysm detection method, apparatus and computer readable storage medium |
CN110211087A (en) * | 2019-01-28 | 2019-09-06 | 南通大学 | The semi-automatic diabetic eyeground pathological changes mask method that can share |
CN111127425A (en) * | 2019-12-23 | 2020-05-08 | 北京至真互联网技术有限公司 | Target detection positioning method and device based on retina fundus image |
CN111341462A (en) * | 2020-02-27 | 2020-06-26 | 上海泗科智能科技有限公司 | Mobile phone terminal diabetic retinopathy screening APP based on deep learning |
CN111951214A (en) * | 2020-06-24 | 2020-11-17 | 北京百度网讯科技有限公司 | Method and device for segmenting readable area in image, electronic equipment and storage medium |
CN112883962A (en) * | 2021-01-29 | 2021-06-01 | 北京百度网讯科技有限公司 | Fundus image recognition method, device, apparatus, storage medium, and program product |
CN115049734A (en) * | 2022-08-12 | 2022-09-13 | 摩尔线程智能科技(北京)有限责任公司 | Method and device for positioning target object in image |
CN116152253A (en) * | 2023-04-20 | 2023-05-23 | 四川大学华西医院 | Cardiac magnetic resonance mapping quantification method, system and storage medium |
CN116703915A (en) * | 2023-08-07 | 2023-09-05 | 天津医科大学眼科医院 | Diabetes retinopathy consultation method and system based on multiple image fusion |
CN116824116A (en) * | 2023-06-26 | 2023-09-29 | 爱尔眼科医院集团股份有限公司 | Super wide angle fundus image identification method, device, equipment and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102087652A (en) * | 2009-12-08 | 2011-06-08 | 百度在线网络技术(北京)有限公司 | Method for screening images and system thereof |
CN103458772A (en) * | 2011-04-07 | 2013-12-18 | 香港中文大学 | Method and device for retinal image analysis |
CN105069803A (en) * | 2015-08-19 | 2015-11-18 | 西安交通大学 | Classifier for micro-angioma of diabetes lesion based on colored image |
CN105761258A (en) * | 2016-02-06 | 2016-07-13 | 上海市第人民医院 | Retinal fundus image bleeding detection method |
CN106530295A (en) * | 2016-11-07 | 2017-03-22 | 首都医科大学 | Fundus image classification method and device of retinopathy |
CN107067409A (en) * | 2017-05-09 | 2017-08-18 | 上海联影医疗科技有限公司 | A kind of blood vessel separation method and system |
CN107480644A (en) * | 2017-08-21 | 2017-12-15 | 吉林大学 | The positioning of optic disk and dividing method, device and storage medium in eye fundus image |
-
2017
- 2017-12-28 CN CN201711456141.4A patent/CN108185984A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102087652A (en) * | 2009-12-08 | 2011-06-08 | 百度在线网络技术(北京)有限公司 | Method for screening images and system thereof |
CN103458772A (en) * | 2011-04-07 | 2013-12-18 | 香港中文大学 | Method and device for retinal image analysis |
CN105069803A (en) * | 2015-08-19 | 2015-11-18 | 西安交通大学 | Classifier for micro-angioma of diabetes lesion based on colored image |
CN105761258A (en) * | 2016-02-06 | 2016-07-13 | 上海市第人民医院 | Retinal fundus image bleeding detection method |
CN106530295A (en) * | 2016-11-07 | 2017-03-22 | 首都医科大学 | Fundus image classification method and device of retinopathy |
CN107067409A (en) * | 2017-05-09 | 2017-08-18 | 上海联影医疗科技有限公司 | A kind of blood vessel separation method and system |
CN107480644A (en) * | 2017-08-21 | 2017-12-15 | 吉林大学 | The positioning of optic disk and dividing method, device and storage medium in eye fundus image |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109472781A (en) * | 2018-10-29 | 2019-03-15 | 电子科技大学 | A kind of diabetic retinopathy detection system based on serial structure segmentation |
CN109472781B (en) * | 2018-10-29 | 2022-02-11 | 电子科技大学 | Diabetic retinopathy detection system based on serial structure segmentation |
CN109544540A (en) * | 2018-11-28 | 2019-03-29 | 东北大学 | A kind of diabetic retina picture quality detection method based on image analysis technology |
CN109829882B (en) * | 2018-12-18 | 2020-10-27 | 广州比格威医疗科技有限公司 | Method for predicting diabetic retinopathy stage by stage |
CN109829882A (en) * | 2018-12-18 | 2019-05-31 | 苏州比格威医疗科技有限公司 | A kind of stages of DR prediction technique |
CN109846513A (en) * | 2018-12-18 | 2019-06-07 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging method, system and image measuring method, processing system and medium |
CN109846513B (en) * | 2018-12-18 | 2022-11-25 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging method, ultrasonic imaging system, image measuring method, image processing system, and medium |
CN110211087B (en) * | 2019-01-28 | 2023-06-30 | 南通大学 | Sharable semiautomatic marking method for diabetic fundus lesions |
CN110211087A (en) * | 2019-01-28 | 2019-09-06 | 南通大学 | The semi-automatic diabetic eyeground pathological changes mask method that can share |
CN109948622B (en) * | 2019-03-26 | 2020-12-11 | 数坤(北京)网络科技有限公司 | Method and device for detecting head and neck body aneurysm and computer readable storage medium |
CN109948622A (en) * | 2019-03-26 | 2019-06-28 | 数坤(北京)网络科技有限公司 | A kind of neck body aneurysm detection method, apparatus and computer readable storage medium |
CN111127425A (en) * | 2019-12-23 | 2020-05-08 | 北京至真互联网技术有限公司 | Target detection positioning method and device based on retina fundus image |
CN111127425B (en) * | 2019-12-23 | 2023-04-28 | 北京至真互联网技术有限公司 | Target detection positioning method and device based on retina fundus image |
CN111341462A (en) * | 2020-02-27 | 2020-06-26 | 上海泗科智能科技有限公司 | Mobile phone terminal diabetic retinopathy screening APP based on deep learning |
CN111951214B (en) * | 2020-06-24 | 2023-07-28 | 北京百度网讯科技有限公司 | Method and device for dividing readable area in image, electronic equipment and storage medium |
CN111951214A (en) * | 2020-06-24 | 2020-11-17 | 北京百度网讯科技有限公司 | Method and device for segmenting readable area in image, electronic equipment and storage medium |
CN112883962A (en) * | 2021-01-29 | 2021-06-01 | 北京百度网讯科技有限公司 | Fundus image recognition method, device, apparatus, storage medium, and program product |
CN112883962B (en) * | 2021-01-29 | 2023-07-18 | 北京百度网讯科技有限公司 | Fundus image recognition method, fundus image recognition apparatus, fundus image recognition device, fundus image recognition program, and fundus image recognition program |
CN115049734A (en) * | 2022-08-12 | 2022-09-13 | 摩尔线程智能科技(北京)有限责任公司 | Method and device for positioning target object in image |
CN116152253A (en) * | 2023-04-20 | 2023-05-23 | 四川大学华西医院 | Cardiac magnetic resonance mapping quantification method, system and storage medium |
CN116152253B (en) * | 2023-04-20 | 2023-06-16 | 四川大学华西医院 | Cardiac magnetic resonance mapping quantification method, system and storage medium |
CN116824116A (en) * | 2023-06-26 | 2023-09-29 | 爱尔眼科医院集团股份有限公司 | Super wide angle fundus image identification method, device, equipment and storage medium |
CN116703915A (en) * | 2023-08-07 | 2023-09-05 | 天津医科大学眼科医院 | Diabetes retinopathy consultation method and system based on multiple image fusion |
CN116703915B (en) * | 2023-08-07 | 2023-09-29 | 天津医科大学眼科医院 | Diabetes retinopathy consultation method and system based on multiple image fusion |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108185984A (en) | The method that eyeground color picture carries out eyeground lesion identification | |
CN108416344B (en) | Method for locating and identifying eyeground color optic disk and yellow spot | |
Qureshi et al. | Combining algorithms for automatic detection of optic disc and macula in fundus images | |
CN107451998B (en) | Fundus image quality control method | |
Rapantzikos et al. | Detection and segmentation of drusen deposits on human retina: Potential in the diagnosis of age-related macular degeneration | |
Bae et al. | A study on hemorrhage detection using hybrid method in fundus images | |
Medhi et al. | An effective fovea detection and automatic assessment of diabetic maculopathy in color fundus images | |
US8098907B2 (en) | Method and system for local adaptive detection of microaneurysms in digital fundus images | |
KR102313143B1 (en) | Diabetic retinopathy detection and severity classification apparatus Based on Deep Learning and method thereof | |
Rasta et al. | Detection of retinal capillary nonperfusion in fundus fluorescein angiogram of diabetic retinopathy | |
Stapor et al. | Segmentation of fundus eye images using methods of mathematical morphology for glaucoma diagnosis | |
CN108230306A (en) | Eyeground color picture blood vessel and arteriovenous recognition methods | |
Li et al. | Vessel recognition of retinal fundus images based on fully convolutional network | |
Vimal et al. | RETRACTED ARTICLE: A method of progression detection for glaucoma using K-means and the GLCM algorithm toward smart medical prediction | |
Mendonça et al. | Segmentation of the vascular network of the retina | |
Acharya et al. | Swarm intelligence based adaptive gamma corrected (SIAGC) retinal image enhancement technique for early detection of diabetic retinopathy | |
Helwan et al. | ISIBC: an intelligent system for identification of breast cancer | |
Sagar et al. | A novel integrated approach using dynamic thresholding and edge detection (idted) for automatic detection of exudates in digital fundus retinal images | |
Bhaskaranand et al. | EyeArt+ EyePACS: automated retinal image analysis for diabetic retinopathy screening in a telemedicine system | |
Brancati et al. | Automatic segmentation of pigment deposits in retinal fundus images of Retinitis Pigmentosa | |
Maheswari et al. | A novel approach for retinal lesion detection in diabetic retinopathy images | |
ManojKumar et al. | Feature extraction from the fundus images for the diagnosis of diabetic retinopathy | |
Jana et al. | A semi-supervised approach for automatic detection and segmentation of optic disc from retinal fundus image | |
CN110930346B (en) | Automatic detection method and storage device for eyeground image microangioma | |
Ahmed et al. | Retina based biometric authentication using phase congruency |
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: 20180622 |