CN110490843A - A kind of eye fundus image blood vessel segmentation method - Google Patents
A kind of eye fundus image blood vessel segmentation method Download PDFInfo
- Publication number
- CN110490843A CN110490843A CN201910667719.3A CN201910667719A CN110490843A CN 110490843 A CN110490843 A CN 110490843A CN 201910667719 A CN201910667719 A CN 201910667719A CN 110490843 A CN110490843 A CN 110490843A
- Authority
- CN
- China
- Prior art keywords
- convolution
- blood vessel
- image
- sampling
- eye fundus
- 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
- 230000011218 segmentation Effects 0.000 title claims abstract description 59
- 210000004204 blood vessel Anatomy 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000005070 sampling Methods 0.000 claims abstract description 43
- 239000000284 extract Substances 0.000 claims abstract description 14
- 230000004913 activation Effects 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims description 18
- 230000008439 repair process Effects 0.000 claims description 5
- 230000007423 decrease Effects 0.000 claims description 4
- 239000008280 blood Substances 0.000 claims description 3
- 210000004369 blood Anatomy 0.000 claims description 3
- 238000012549 training Methods 0.000 abstract description 17
- 230000002792 vascular Effects 0.000 abstract description 5
- 230000000694 effects Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 206010012601 diabetes mellitus Diseases 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 206010020772 Hypertension Diseases 0.000 description 1
- 208000034189 Sclerosis Diseases 0.000 description 1
- 210000002565 arteriole Anatomy 0.000 description 1
- 210000001367 artery Anatomy 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 210000000416 exudates and transudate Anatomy 0.000 description 1
- 208000030533 eye disease Diseases 0.000 description 1
- 210000004220 fundus oculi Anatomy 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 201000011066 hemangioma Diseases 0.000 description 1
- 230000001631 hypertensive effect Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 210000001210 retinal vessel Anatomy 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 210000001835 viscera Anatomy 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
-
- 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
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Abstract
The invention discloses a kind of eye fundus image blood vessel segmentation methods, it include: that (a) is sampled in original image in the form of sliding window, to expand eyeground data, and training is done with the picture in window, the contextual information of center pixel vertex neighborhood is contained in window;(b) bilinearity difference is taken to up-sample in convolutional layer the image of input, by the image after up-sampling, local fine feature is extracted by common convolution, extracting the global characteristics under large scale by global convolution and expansion convolution and using step-length is 2 convolutional layer by scaling;(c) it then carries out carrying out down-sampling using the operation in convolution and maximum pond and extracts its abstract characteristics, dense connection is carried out to this structure on same scale;(d) by the way that batch normalized function and activation primitive are added before convolutional layer each in block structure, the segmentation to fine vascular is realized.
Description
Technical field
The present invention relates to image partition method, in particular to a kind of eye fundus image blood vessel segmentation method.
Background technique
The important ring that eye fundus image analysis is doctor during fundus oculi disease diagnosis, not only with eye disease phase
The case where closing, but also can reflect other diseases and internal organs blood vessel [1].Therefore the variation of eye fundus image, to a certain extent secretly
Show that the state of body intracorporeal organ is changed.As the retinal vessel of hypertensive patient is likely to occur the feelings of artery sclerosis
Condition, diabetic eyeground are likely to occur the focal areas such as arteriole hemangioma [2], this just reflects whole body to a certain extent
The change situation of blood vessel, doctor can analyze accordingly, judge the severity of disease.In the analytic process of eye fundus image, need
Fine segmentation is carried out to blood vessel, and since there may be blutpunkte, exudate and focal area, blood vessels for eye fundus image
Itself the problems such as there are the fine vasculars of center line reflection and low contrast, so that the blood vessel fining segmentation of eye fundus image is always
It is all the hot issue studied both at home and abroad.
In recent years, in eye fundus image blood vessel segmentation field, domestic and foreign scholars are done by traditional images Processing Algorithm with supervision
A series of achievement is gone out.In traditional images Processing Algorithm research, emerged a collection of new algorithm, as matched filter [3],
The advantages of optical fundus blood vessel partitioning algorithm [4] based on template etc., such algorithm is without training sample, according to the line of eye fundus image
It is stronger that the features such as reason, color are split interpretation, but its precision is lower, and the accuracy rate of minute blood vessel segmentation is lower.And it supervises
It superintends and directs method and mainly passes through extraction feature, and feature is sent into classifier to realize the segmentation of blood vessel.Most of supervision algorithm at present
Be by manually extracting feature, such as Ricci [5] fixed using length but two orthogonal thread detectors of direction change come
Statistics is located at the average gray value on vessel segment, and the feature vector of extraction is sent to support vector machines (Support
Vector machine, SVM) in carry out blood vessel segmentation.
It has been proposed that a kind of supervision algorithm based on decision tree, by using gradient vector field, Morphological scale-space, line feature
Feature vector is extracted with 4 kinds of methods such as Gabor responses, and is sent into decision tree and classifies.With traditional images processing method phase
Than the segmentation effect of measure of supervision increases.But it manually extracts feature and depends on priori knowledge [7], and characteristic Design
There are limitations.In order to solve the limitation of manual features, there is scholar to automatically extract blood by projected depth learning network model
Pipe feature is simultaneously classified.
In addition, optical fundus blood vessel segmentation is regarded as a border detection problem, by by full convolutional neural networks and Quan Lian
It connects condition random field (Conditional random field, CRF) to combine, blood vessel is split.Orlando etc. is proposed
A kind of special CRF model is split blood vessel, and CRF can learn consistency of the image in structure, but the impression of CRF
Open country is limited can not to extract global semantic information.The it is proposeds such as Liskowski are split optical fundus blood vessel using convolutional neural networks,
The method regarded blood vessel segmentation problem as two classification problems, and propose the variant and integrated structureization prediction of many networks,
Allow to simultaneously classify to multiple pixels.Supervision algorithm attempts to look for the mapping function inputted between output
[11], training pattern is come by using the standard picture of expert's manual segmentation.
But the blood vessel segmentation model based on convolutional neural networks proposed at present still cannot identify well it is tiny
Blood vessel, and there are noises in segmentation result.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of eye fundus image blood vessel segmentation methods, can be to the image of input
It takes bilinearity difference to be up-sampled in convolutional layer, by the image after up-sampling, local fine is extracted by common convolution
Change feature;It then carries out carrying out down-sampling using the operation in convolution and maximum pond and extracts its abstract characteristics, in same scale
On dense connection is carried out to this structure, strengthen high dimensional feature and merged with the information of low-dimensional feature, to be better achieved pair
The segmentation of fine vascular;It realizes the optimization to network and further increases the generalization ability of network.
In order to solve the above-mentioned technical problem, the technical solution of the present invention is as follows:
A kind of eye fundus image blood vessel segmentation method, comprising:
(a) it is sampled in original image in the form of sliding window, to expand eyeground data, and with the picture in window
Training is done, the contextual information of center pixel vertex neighborhood is contained in window;
(b) it takes bilinearity difference to up-sample in convolutional layer the image of input, the image after up-sampling leads to
It crosses common convolution and extracts local fine feature, the global characteristics under large scale are extracted simultaneously by global convolution and expansion convolution
Using step-length is 2 convolutional layer by scaling;
(c) it then carries out carrying out down-sampling using the operation in convolution and maximum pond and extracts its abstract characteristics, same
Dense connection is carried out to this structure on scale;
(d) it by the way that batch normalized function and activation primitive are added before convolutional layer each in block structure, realizes to micro-
The segmentation of thin blood vessel.
Preferably, semantic segmentation network implementations of the eye fundus image blood vessel segmentation method based on Pixel-level, wherein described
Semantic segmentation network is made of constricted path and path expander, and wherein constricted path gradually decreases image by down-sampling operation
Spatial Dimension, and path expander gradually repairs the details and Spatial Dimension of object by up-sampling operation.
Preferably, the sampling in step (a) the following steps are included:
(a1) Ultrafine feature extraction is carried out, that is, carries out 1 up-sampling, the dimension of picture for inputting network access network is enlarged into originally
2 times;
(a2) feature extraction is carried out on this basis and former input size is zoomed to by the convolution kernel that step-length is 2,
(a3) by 2 down-samplings, 2 up-samplings, and constricted path is subjected to jump with path expander and is connected.
Preferably, feature extraction uses the concatenated iteration with different expansion rates and convolution kernel size in step (b)
Convolutional layer carries out feature extraction, and the expansion rate of this two layers of iterative convolution layer is respectively 2 and 3, and convolution kernel size is respectively (3 × 3)
(5 × 5).
By adopting the above technical scheme, the beneficial effects of the present invention are:
Blood vessel segmentation network (semantic segmentation network Vessel SegNet, VSNet) is proposed, from the angle that fine granularity is divided
Degree, improves the segmentation precision of fine vascular.It is sampled in original image in the form of sliding window, to expand eyeground data,
And training is done with the picture in window, the contextual information of center pixel vertex neighborhood is contained in window;
Secondly, the image to input takes bilinearity difference to up-sample in convolutional layer, by the image after up-sampling,
Local fine feature is extracted by common convolution;
It then carries out carrying out down-sampling using the operation in convolution and maximum pond and extracts its abstract characteristics, in same scale
On dense connection is carried out to this structure, strengthen high dimensional feature and merged with the information of low-dimensional feature, to be better achieved pair
The segmentation of fine vascular;
It realizes the optimization to network and further increases the generalization ability of network.
Detailed description of the invention
Fig. 1 is the global convolutional coding structure figure in the method for the present invention;
Fig. 2 is the expansion convolutional coding structure figure in the method for the present invention;
Fig. 3 is the semantic segmentation network diagram in the method for the present invention;
Fig. 4 is point of the eye fundus image dividing method based on semantic segmentation network in the method for the present invention and current method
Cut effect difference signal.
Specific embodiment
Specific embodiments of the present invention will be further explained with reference to the accompanying drawing.It should be noted that for
The explanation of these embodiments is used to help understand the present invention, but and does not constitute a limitation of the invention.In addition, disclosed below
The each embodiment of the present invention involved in technical characteristic can be combined with each other as long as they do not conflict with each other.
The present invention provides a kind of eye fundus image blood vessel segmentation methods, wherein in the eye fundus image blood vessel segmentation method
Image data derives from eye fundus image database: DRIVE and STARE.The paper delivered at present in optical fundus blood vessel segmentation field is more
Using database disclosed in these come the performance of verification algorithm.Wherein DRIVE eye fundus image database is specifically for comparison blood
Pipe partitioning algorithm effect and establish, the eye fundus image in the library is both from diabetic eyeground pathological changes screening tissue.DRIVE number
It is 565 pixels × 584 pixels eye fundus image according to 40 width resolution ratio are shared in library, image is jpeg format.This research is by STARE
In total data and DRIVE database in training set collectively as the training set of VSNet model, will be in DRIVE database
Test set tested as the test set of model, wherein in training set comprising original image, expert's manual segmentation standard
Blood-vessel image, the equal 40 width image of each section are used for network training.Test set each section includes 20 width images, for network
Performance is assessed, and carries out original number using random gamma transformation, in the form of Random-Rotation, random Gaussian be fuzzy and random noise
According to expansion.For the performance for further increasing fine granularity segmentation, this research samples to carry out network training using sliding window.It will
Image after expansion in training set uses the sliding window of 48*48, carries out continuous sampling with 5 for step-length, extracts from training set
Image block sum be 330000.
Semantic segmentation network implementations of the eye fundus image blood vessel segmentation method based on Pixel-level, wherein the semantic segmentation
Network is made of constricted path and path expander, and wherein constricted path gradually decreases the space dimension of image by down-sampling operation
Degree, and path expander gradually repairs the details and Spatial Dimension of object by up-sampling operation.
The semantic segmentation network is made of constricted path and path expander, wherein constricted path by down-sampling operate by
Decrescence lack the Spatial Dimension of image, and path expander gradually repairs the details and Spatial Dimension of object by up-sampling operation.Its
Network structure uses the structure of SFConv to carry out Ultrafine feature extraction first, that is, carries out 1 up-sampling, will input the figure of network access network
Chip size is enlarged into original 2 times, uses GCN and IDCN to carry out feature extraction on this basis and passes through step-length as 2 convolution
Core zooms to former input size, later by 2 down-samplings, 2 up-samplings, and constricted path and path expander is jumped
Connection.
As shown in Figure 1, the structure of global convolution GCN, which employs the schemes that two-way convolution is spliced parallel, wherein all the way
Convolution kernel is the form of (1 × N) and (N × 1), and the convolution kernel of another way is the form of (N × 1) and (1 × N), and this design is thought
Road can sufficiently expand receptive field under the premise of not increasing quantity of parameters.Pass through lot of experiment validation herein, when N=8
Effect is best.
As shown in Fig. 2, the structure of expansion convolution IDCNN, using different expansion rates, to expand the hole in convolution kernel
To under the premise of not increasing parameter amount, increase convolution kernel receptive field, there is employed herein it is concatenated have different expansion rates with
The IDCNN of convolution kernel size carries out feature extraction, the expansion rate of this two layers of IDCNN is respectively 2 and 3, and convolution kernel size is respectively
(3 × 3) and (5 × 5), so as to avoid characteristic loss brought by pinhole.
If carrying out multiple down-sampling to image block and up-sampling operating, it will lead to the location information of blood vessel feature and specific
Loss in detail is serious, therefore only 2 up-samplings avoid the loss in detail of feature with down-sampling herein.
In order to make full use of the characteristic information that each convolutional layer extracts in network, this research is by the way of jump connection, i.e.,
The characteristic pattern of same scale is subjected to splicing fusion, the characteristic pattern to reduce the loss of feature, then after merging carries out convolution to mention
The utilization rate of high feature.Finally the convolution kernel of (1 × 1) is used to carry out feature extraction to image block, can effectively reduced trained
Parameter simultaneously accelerates inference speed.
The essence of full convolutional neural networks training is the rule [15] learnt between image block.Since the image of every batch of training is small
Block all randomly selects, then network just needs to go to learn different distributions every time, it is slack-off to will lead to training speed in this way.Together
When, in order to utilize the nonlinear characteristic of activation primitive, the distribution between data cannot be excessive.In activation primitive and convolutional layer
It joined batch normalized function (BN) before, the image block of every batch of training done into normalized.
Semantic segmentation network diagram as shown in Figure 3, wherein the corresponding multi-channel feature figure of each bar-shaped frame;
The bar-shaped frame of green is the characteristic pattern of duplication;Arrow indicates to carry out different operations.Semantic segmentation network Vessel SegNet packet
Containing constricted path and path expander two parts, shrink, path expander respectively includes 2 block structures and symmetrical.Constricted path
Middle block structure is that the pondization operation for being 2 by step-length is connected, therefore the size of input picture is previous in the 2nd block structure
The half of block structure.It is connected between block structure in path expander by up-sampling operation, due to path expander and shrinks road
Diameter is symmetrical, therefore the size of image block and the quantity of convolution kernel are corresponding with constricted path.Each block structure is by 23
× 3 convolutional layer composition, the input by the output of convolutional layer each in block structure as all convolutional layers behind, and pass through
The mode of jump connection is respectively calculated output of the last superposition as the block structure.Make before carrying out convolution operation every time
Network is advanced optimized with ReLU activation primitive and batch normalization layer.Since semantic segmentation network Vessel SegNet is
Network structure end to end, therefore after the feature feeding Sigmoid classifier that network is extracted, after can directly exporting segmentation
Blood vessel probability graph.
As shown in figure 4, through the above, about improving 3% compared with the current the best way of sensitivity.In
The blood vessel segmentation effect obtained on DRIVE eye fundus image database is as shown below, and the 1st row image is taken from DRIVE number in figure
According to the eye fundus image in library, the 2nd row is the vessel graph that expert is divided by hand, and the 3rd row is that the present invention proposes Vessel SegNet net
The blood vessel probability graph of network segmentation.As can be seen that Vessel SegNet network can accurately be partitioned into blood vessel structure, and micro-
Do well in the segmentation of thin blood vessel.
According to above content, the invention proposes a kind of eye fundus image blood vessel segmentation methods, comprising:
(a) it is sampled in original image in the form of sliding window, to expand eyeground data, and with the picture in window
Training is done, the contextual information of center pixel vertex neighborhood is contained in window;
(b) it takes bilinearity difference to up-sample in convolutional layer the image of input, the image after up-sampling leads to
It crosses common convolution and extracts local fine feature, the global characteristics under large scale are extracted simultaneously by global convolution and expansion convolution
Using step-length is 2 convolutional layer by scaling;
(c) it then carries out carrying out down-sampling using the operation in convolution and maximum pond and extracts its abstract characteristics, same
Dense connection is carried out to this structure on scale;
(d) it by the way that batch normalized function and activation primitive are added before convolutional layer each in block structure, realizes to micro-
The segmentation of thin blood vessel.
Semantic segmentation network implementations of the eye fundus image blood vessel segmentation method based on Pixel-level, wherein the semantic segmentation
Network is made of constricted path and path expander, and wherein constricted path gradually decreases the space dimension of image by down-sampling operation
Degree, and path expander gradually repairs the details and Spatial Dimension of object by up-sampling operation.
Specifically, the sampling in step (a) the following steps are included:
(a1) Ultrafine feature extraction is carried out, that is, carries out 1 up-sampling, the dimension of picture for inputting network access network is enlarged into originally
2 times;
(a2) feature extraction is carried out on this basis and former input size is zoomed to by the convolution kernel that step-length is 2,
(a3) by 2 down-samplings, 2 up-samplings, and constricted path is subjected to jump with path expander and is connected.
Preferably, feature extraction uses the concatenated iteration with different expansion rates and convolution kernel size in step (b)
Convolutional layer carries out feature extraction, and the expansion rate of this two layers of iterative convolution layer is respectively 2 and 3, and convolution kernel size is respectively (3 × 3)
(5 × 5).
In conjunction with attached drawing, the embodiments of the present invention are described in detail above, but the present invention is not limited to described implementations
Mode.For a person skilled in the art, in the case where not departing from the principle of the invention and spirit, to these embodiments
A variety of change, modification, replacement and modification are carried out, are still fallen in protection scope of the present invention.
Claims (4)
1. a kind of eye fundus image blood vessel segmentation method characterized by comprising
(a) it is sampled in original image in the form of sliding window, to expand eyeground data, and is instructed with the picture in window
Practice, contains the contextual information of center pixel vertex neighborhood in window;
(b) bilinearity difference is taken to up-sample in convolutional layer the image of input, by the image after up-sampling, by general
Logical convolution extracts local fine feature, extracts global characteristics and the use under large scale by global convolution and expansion convolution
The convolutional layer that step-length is 2 is by scaling;
(c) it then carries out carrying out down-sampling using the operation in convolution and maximum pond and extracts its abstract characteristics, in same scale
On dense connection is carried out to this structure;
(d) it by the way that batch normalized function and activation primitive are added before convolutional layer each in block structure, realizes to fine blood
The segmentation of pipe.
2. eye fundus image blood vessel segmentation method according to claim 1, which is characterized in that the eye fundus image blood vessel segmentation
Semantic segmentation network implementations of the method based on Pixel-level, wherein the semantic segmentation network is by constricted path and path expander group
At wherein constricted path gradually decreases the Spatial Dimension of image by down-sampling operation, and path expander passes through up-sampling operation
Gradually repair the details and Spatial Dimension of object.
3. eye fundus image blood vessel segmentation method according to claim 1, which is characterized in that the sampling packet in step (a)
Include following steps:
(a1) Ultrafine feature extraction is carried out, that is, carries out 1 up-sampling, the dimension of picture for inputting network access network is enlarged into original 2
Times;
(a2) feature extraction is carried out on this basis and former input size is zoomed to by the convolution kernel that step-length is 2,
(a3) by 2 down-samplings, 2 up-samplings, and constricted path is subjected to jump with path expander and is connected.
4. eye fundus image blood vessel segmentation method according to claim 1, which is characterized in that the feature extraction in step (b)
Feature extraction, this two layers of iterative convolution layer are carried out using the concatenated iterative convolution layer with different expansion rates and convolution kernel size
Expansion rate be respectively 2 and 3, convolution kernel size is respectively (3 × 3) and (5 × 5).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910667719.3A CN110490843A (en) | 2019-07-23 | 2019-07-23 | A kind of eye fundus image blood vessel segmentation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910667719.3A CN110490843A (en) | 2019-07-23 | 2019-07-23 | A kind of eye fundus image blood vessel segmentation method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110490843A true CN110490843A (en) | 2019-11-22 |
Family
ID=68548040
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910667719.3A Pending CN110490843A (en) | 2019-07-23 | 2019-07-23 | A kind of eye fundus image blood vessel segmentation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110490843A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112070658A (en) * | 2020-08-25 | 2020-12-11 | 西安理工大学 | Chinese character font style migration method based on deep learning |
CN112734784A (en) * | 2021-01-28 | 2021-04-30 | 依未科技(北京)有限公司 | High-precision fundus blood vessel boundary determining method, device, medium and equipment |
CN112967209A (en) * | 2021-04-23 | 2021-06-15 | 上海大学 | Endoscope image blood vessel texture enhancement method based on multiple sampling |
CN113256662A (en) * | 2021-06-28 | 2021-08-13 | 杭州迪英加科技有限公司 | Pathological section image segmentation method and device, computer equipment and storage medium |
CN112070658B (en) * | 2020-08-25 | 2024-04-16 | 西安理工大学 | Deep learning-based Chinese character font style migration method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20170113251A (en) * | 2016-03-24 | 2017-10-12 | 재단법인 아산사회복지재단 | Method and device for automatic inner and outer vessel wall segmentation in intravascular ultrasound images using deep learning |
CN107909585A (en) * | 2017-11-14 | 2018-04-13 | 华南理工大学 | Inner membrance dividing method in a kind of blood vessel of intravascular ultrasound image |
CN109671094A (en) * | 2018-11-09 | 2019-04-23 | 杭州电子科技大学 | A kind of eye fundus image blood vessel segmentation method based on frequency domain classification |
-
2019
- 2019-07-23 CN CN201910667719.3A patent/CN110490843A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20170113251A (en) * | 2016-03-24 | 2017-10-12 | 재단법인 아산사회복지재단 | Method and device for automatic inner and outer vessel wall segmentation in intravascular ultrasound images using deep learning |
CN107909585A (en) * | 2017-11-14 | 2018-04-13 | 华南理工大学 | Inner membrance dividing method in a kind of blood vessel of intravascular ultrasound image |
CN109671094A (en) * | 2018-11-09 | 2019-04-23 | 杭州电子科技大学 | A kind of eye fundus image blood vessel segmentation method based on frequency domain classification |
Non-Patent Citations (1)
Title |
---|
ALI HATAMIZADEH: "Deep Dilated Convolutional Nets for the Automatic Segmentation of Retinal Vessels", 《ARXIV:1905.12120》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112070658A (en) * | 2020-08-25 | 2020-12-11 | 西安理工大学 | Chinese character font style migration method based on deep learning |
CN112070658B (en) * | 2020-08-25 | 2024-04-16 | 西安理工大学 | Deep learning-based Chinese character font style migration method |
CN112734784A (en) * | 2021-01-28 | 2021-04-30 | 依未科技(北京)有限公司 | High-precision fundus blood vessel boundary determining method, device, medium and equipment |
CN112967209A (en) * | 2021-04-23 | 2021-06-15 | 上海大学 | Endoscope image blood vessel texture enhancement method based on multiple sampling |
CN113256662A (en) * | 2021-06-28 | 2021-08-13 | 杭州迪英加科技有限公司 | Pathological section image segmentation method and device, computer equipment and storage medium |
CN113256662B (en) * | 2021-06-28 | 2021-11-05 | 杭州迪英加科技有限公司 | Pathological section image segmentation method and device, computer equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Alyoubi et al. | Diabetic retinopathy detection through deep learning techniques: A review | |
Soomro et al. | Strided fully convolutional neural network for boosting the sensitivity of retinal blood vessels segmentation | |
CN110097559B (en) | Fundus image focus region labeling method based on deep learning | |
CN112132817B (en) | Retina blood vessel segmentation method for fundus image based on mixed attention mechanism | |
CN109345538A (en) | A kind of Segmentation Method of Retinal Blood Vessels based on convolutional neural networks | |
CN111046835A (en) | Eyeground illumination multiple disease detection system based on regional feature set neural network | |
CN110490843A (en) | A kind of eye fundus image blood vessel segmentation method | |
CN114066884B (en) | Retinal blood vessel segmentation method and device, electronic device and storage medium | |
CN115205300A (en) | Fundus blood vessel image segmentation method and system based on cavity convolution and semantic fusion | |
Afolabi et al. | The use of U-Net lite and Extreme Gradient Boost (XGB) for glaucoma detection | |
CN110689526A (en) | Retinal blood vessel segmentation method and system based on retinal fundus image | |
CN115375711A (en) | Image segmentation method of global context attention network based on multi-scale fusion | |
Ren et al. | An improved U-net based retinal vessel image segmentation method | |
CN115661066A (en) | Diabetic retinopathy detection method based on segmentation and classification fusion | |
Pavithra et al. | Computer aided diagnosis of diabetic macular edema in retinal fundus and OCT images: A review | |
HaoQi et al. | CGAN-based synthetic medical image augmentation between retinal fundus images and vessel segmented images | |
Qin et al. | A review of retinal vessel segmentation for fundus image analysis | |
Mansour | Using genetic algorithm for identification of diabetic retinal exudates in digital color images | |
Miao et al. | Classification of Diabetic Retinopathy Based on Multiscale Hybrid Attention Mechanism and Residual Algorithm | |
Liu et al. | Feature pyramid U‐Net for retinal vessel segmentation | |
Tian et al. | Learning discriminative representations for fine-grained diabetic retinopathy grading | |
Cao et al. | Attentional mechanisms and improved residual networks for diabetic retinopathy severity classification | |
Shenkut et al. | Fundus GAN-GAN-based fundus image synthesis for training retinal image classifiers | |
Li et al. | MC-UNet: Multimodule Concatenation Based on U-Shape Network for Retinal Blood Vessels Segmentation | |
CN115619814A (en) | Method and system for jointly segmenting optic disk and optic cup |
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: 20191122 |