CN106408564A - Depth-learning-based eye-fundus image processing method, device and system - Google Patents
Depth-learning-based eye-fundus image processing method, device and system Download PDFInfo
- Publication number
- CN106408564A CN106408564A CN201610884671.8A CN201610884671A CN106408564A CN 106408564 A CN106408564 A CN 106408564A CN 201610884671 A CN201610884671 A CN 201610884671A CN 106408564 A CN106408564 A CN 106408564A
- Authority
- CN
- China
- Prior art keywords
- image
- deep learning
- eye fundus
- fundus image
- resampling
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- 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
Abstract
The application discloses a depth-learning-based eye-fundus image processing method, device and system. The method comprises: eye-fundus image pretreatment is carried out and region segmentation extraction is carried out on the eye-fundus image; resampling is carried out on a regional image obtained by the segmentation extraction; data expansion is carried out on the regional image after resampling; and a regional image feature is identified by using a depth learning method. Therefore, an eye-fundus image can be analyzed automatically; the analysis result is objective and accurate; and the manpower cost is saved.
Description
Technical field
The present invention relates to eye fundus image process field, process in particular to a kind of eye fundus image based on deep learning
Method and system.
Background technology
The analysis of eye fundus image is predominantly manually carried out in the prior art, because artificial analysis is with stronger subjectivity
Property, therefore the accuracy of people's manual analyses and concordance are difficult to be ensured well.Human cost also more and more higher simultaneously, because
This automatic analysis technology has very big advantage as more objective and human cost saving aspect.
The concept of deep learning comes from the research of artificial neural network.Multilayer perceptron containing many hidden layers is exactly a kind of depth
Learning structure.Deep learning is passed through to combine the more abstract high-rise expression attribute classification of low-level feature formation or feature, to find
The distributed nature of data represents.The concept of deep learning is proposed in 2006 by Hinton et al..Based on depth confidence network
(DBN) propose non-supervisory greed successively training algorithm, bring hope for solving the related optimization difficult problem of deep structure, subsequently propose
Multilamellar autocoder deep structure.In addition the convolutional neural networks that Lecun et al. proposes are first real multiple structures
Practise algorithm, its utilization space relativeness reduces number of parameters to improve training performance.Deep learning is in machine learning research
A new field, its motivation be set up, simulation human brain be analyzed learn neutral net, it imitate human brain mechanism
To explain data, such as image, sound and text.
Deep learning technology has application in a lot of fields, also result in certain attention in field of medical images, replaces
Artificial experience medical imaging analysis feature, deep learning technology can automatically learn view data abstract spy at all levels
Levy, thus the substitutive characteristics of preferably response data, reduce the incompleteness that artificial design feature causes.
Although deep learning can automatically learning model feature it is possible to reach good accuracy of identification, a kind of
Data driven type model, needs the data of " quite big " magnitude, and simulation human brain vision mechanism automatically learns to each layer of data
Secondary abstract characteristics, thus preferably reflect the essence of data.Typically first in the abstract characteristics of image of pixel layer, to each pixel mark
Remember its affiliated species, final output result is exactly the identification of feature, contain taking 2048 × 2048 eye fundus image as a example
More than 4,000,000 pixels, the amount of calculation finding characteristics of image in pixel scale is extremely huge, and the time of this algorithm is multiple
Miscellaneous degree is drastically lifted, and in order to ensure algorithm real-time, some deep learning algorithms are contracted to image in preprocessing process
Little, to improve calculating speed.Image reduce the fusion inevitably resulting in pixel and surrounding pixel, and then have impact on and with pixel be
Rank selects the accuracy of characteristics of image.As not downscaled images, ensure the accuracy calculating simultaneously, then need higher parallel volume
Journey skill and more preferably more hardware supported, otherwise calculating speed is very slow.Due to this temporal inadmissibility, limit
Make the popularization in practical application for the deep learning.
Content of the invention
The subject matter that the application solves is to provide a kind of method for processing fundus images based on deep learning, device and be
System, manually carries out the analysis of eye fundus image, analysis result subjective human cost simultaneously also more and more higher in prior art
Technical problem.
In order to solve above-mentioned technical problem, the invention discloses a kind of method for processing fundus images based on deep learning its
It is characterised by, comprise the following steps：Eye fundus image pretreatment first, carries out the segmented extraction in region to eye fundus image；Then right
The area image that segmented extraction obtains carries out resampling；Data extending is carried out to the area image of resampling；Using deep learning
Method identification region characteristics of image.
Preferably, described eye fundus image pretreatment also includes calculating whole eye fundus image field color meansigma methodss, optical fundus figure
As any pixel deducts this meansigma methodss color.
Preferably, described region includes optic disc region, optic cup region, circumpapillary atrophy region and angiosomeses.
Preferably, described resampling includes for area image generating fettucelle and cut out areas image.
Preferably, described data extending includes area image scaling, area image rotation, area image inclination and/or area
The regulation of area image contrast.
Preferably, described area image features training and region are included using deep learning method identification region characteristics of image
The identification of characteristics of image.
Preferably, described deep learning method is convolutional neural networks method.
The invention also discloses a kind of eye fundus image processing meanss based on deep learning are it is characterised in that include：
Eye fundus image pretreatment module, described for pretreatment is carried out to eye fundus image, region is carried out to eye fundus image
Segmented extraction；
Resampling module, the area image that described resampling module is used for segmented extraction is obtained carries out resampling；
Data extending module, described data extending module carries out data extending to the area image of resampling；And
Area image feature recognition module, described area image feature recognition module is used for using the identification of deep learning method
Area image feature.
The invention also discloses a kind of eye fundus image processing system based on deep learning is it is characterised in that include optical fundus
Image-input device, described eye fundus image input equipment is connected with eye fundus image processing meanss, described eye fundus image processing meanss
It is connected with result output device.
Compared with prior art, a kind of method for processing fundus images based on deep learning described herein, device and
System, has reached following effect：
(1) method for processing fundus images based on deep learning in the application, apparatus and system, it is possible to achieve automatically right
Eye fundus image is analyzed, and analysis result is objective and accurate.
(2) method for processing fundus images based on deep learning in the application, apparatus and system, using deep learning
Method, automatically study view data abstract characteristics at all levels, thus the substitutive characteristics of preferably response data, reduce people
The incompleteness causing for design feature.
(3) method for processing fundus images based on deep learning in the application, apparatus and system, in Image semantic classification mistake
Journey adds optic disc, optic cup, circumpapillary atrophy region, the segmented extraction of angiosomeses, as the data of subsequent deep learning.
So neither need downscaled images, also can complete the training of data in the acceptable time, solve deep learning in time
The problem of inadmissibility.
Brief description
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the present invention, this
Bright schematic description and description is used for explaining the present invention, does not constitute inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 is the flow chart of the method for processing fundus images based on deep learning described in the embodiment of the present invention one；
Fig. 2 is the employing convolutional Neural of the method for processing fundus images based on deep learning described in the embodiment of the present invention three
Network training process flow chart；
Fig. 3 is the structured flowchart of the eye fundus image processing meanss based on deep learning described in the embodiment of the present invention four；
Fig. 4 is the structured flowchart of the eye fundus image processing system based on deep learning described in the embodiment of the present invention five；
Fig. 5 is the flow chart of the eye fundus image preprocess method described in the embodiment of the present invention five；
Fig. 6 be the optical fundus blood vessel of eye fundus image preprocess method described in the embodiment of the present invention five parabola aspect graph and
The direction collection of illustrative plates of optical fundus blood vessel；
Fig. 7 is that the optical fundus blood vessel trend of the eye fundus image preprocess method described in the embodiment of the present invention five is indicated method
Procedure chart；
Fig. 8 is the optic disc under different brightness in eye fundus image preprocess method described in the embodiment of the present invention five；
Fig. 9 be the middle determination optic disc area of the eye fundus image preprocess method described in the embodiment of the present invention five and macular area figure with
Corresponding blood vessel distance field figure.
Specific embodiment
To censure specific components as employed some vocabulary in the middle of description and claim.Those skilled in the art should
It is understood that hardware manufacturer may call same assembly with different nouns.This specification and claims are not with name
The difference claiming is used as distinguishing the mode of assembly, but difference functionally is used as the criterion distinguished with assembly.As logical
In the middle of piece description and claim, mentioned "comprising" is an open language, therefore should be construed to " comprise but do not limit
In "." substantially " refer in receivable range of error, those skilled in the art can solve described in the range of certain error
Technical problem, basically reaches described technique effect.Description subsequent descriptions are to implement the better embodiment of the application, so described
Description is for the purpose of the rule that the application is described, is not limited to scope of the present application.The protection domain of the application
When being defined depending on the defined person of claims.
Below in conjunction with accompanying drawing, the application is described in further detail, but not as the restriction to the application.Embodiment one
As shown in figure 1, for the method for processing fundus images based on deep learning in the present embodiment it is characterised in that wrapping
Include step S100 to step S400：
S100：Eye fundus image pretreatment first, carries out the segmented extraction in region to eye fundus image；
S200：Then area image segmented extraction being obtained carries out resampling；
S300：Data extending is carried out to the area image of resampling；
S400：Using deep learning method identification region characteristics of image.
Embodiment two
On the basis of embodiment one, wherein step S100 comprises the following steps：First, in order to eliminate due to illumination bar
Difference between the image that part, the difference of camera resolution cause, calculates whole eye fundus image field color meansigma methodss, optical fundus figure
As any pixel deducts this meansigma methodss color.Secondly, using multi-level parabola, optical fundus blood vessel form is described, identifies eye
The centrage of bottom blood vessel.Then, move towards to obtain optic disc regional location (place that blood vessel is assembled) using blood vessel, use ellipse fitting side
Formula obtains optic disc region.With optic disc center as initial point, from initial point to optic disc, farthest border is in the outside distance expanding 50 pixels
For radius, extract optic cup region and circumpapillary atrophy region (PPA).
Preferably, step S200 comprises the following steps：Extract image in regenerate multiple sizes be 128 pixels ×
The square dough sheet of 128 pixels (alternatively other sizes), two neighboring square center distance is 32 pixels.In addition, being
Remove boundary effect, cut out the image of resampling to full-sized 90%.
Preferably, step S300 comprises the following steps：Random zoomed image ± 10%, spends scope inward turning 0 to 360 at random
Turn image, (rad is angular unit to random inclination ± 0.2rad：Radian)；The tune for areal image contrast can also be included
Saving to increase the quantity of training；By the parameter value of modified-image, obtain substantial amounts of view data so that the quantity of area image
Increase, thus reaching the purpose of expanding data.
Embodiment three
As shown in Fig. 2 being using convolutional neural networks training process flow chart, carry out pre- place firstly for eye fundus image
Reason, then carries out resampling data expansion according to the method in embodiment two, the convolutional neural networks after training are to optical fundus figure
As being identified and analyzed.
As shown in Fig. 2 convolutional neural networks framework includes 5 convolutional layers with weight and 2 full articulamentums, input
The image that layer generates for resampling in Image semantic classification step.Be connected with input layer is 5 convolutional layers
(Convolutional Layers).1st and the 2nd convolutional layer convolution complete after employ ReLUS (rectified linear
Units) function is processed, to accelerate the training speed of neutral net；Then carried out local regularization (Local again
Response Normalization) (formula 1), to prevent over-fitting, finally also carry out maximum pond (MaxPooling).
GivenRepresent the neuron activity of the point (x, y) of application convolution kernel i, this neuron activity is in application ReLUS
Afterwards, progressive regularization activityExpression formula be：
Wherein, n is the adjacent convolution check figure in the same space position, and N is the quantity of all convolution kernels of this convolutional layer, k, n, α,
β is hyper parameter.
The output in maximum pond connects the 3rd convolutional layer as input.3rd, 4,5 convolutional layers are connected with each other, not just
Then change layer and maximum pond layer is coupled.Convolutional layer is connected with two full articulamentums (FullyConnection Layers),
, using the output of the 5th convolutional layer as input, the output of the 2nd full articulamentum is defeated as softmax layer for 1st full articulamentum
Enter, softmax layer is output as the labeled bracketing of characteristics of image automatically extracting.Intend to reduce crossing in full articulamentum
Close, the method that we employ " dropout ", in each training batch, allow a certain proportion of full articulamentum nodal value be 0
Reduce Expired Drugs.
In the present embodiment the method for deep learning we use convolutional neural networks, other nerve net may also be employed
Network method, or additive method, such as the deep learning method based on autocoder combination or restriction Boltzmann machine.
Example IV
As shown in figure 3, being the structured flowchart of the eye fundus image processing meanss based on deep learning, wherein eye fundus image is processed
Device 2 includes eye fundus image pretreatment module 10, resampling module 20, data extending module 30 and area image feature identification mould
Block 40；Wherein eye fundus image pretreatment module 10, resampling module 20, data extending module 30 and area image feature identification mould
Block 40 is linked in sequence；
Eye fundus image pretreatment module 10 is used for carrying out pretreatment to eye fundus image, and eye fundus image is carried out with the segmentation in region
Extract；
The area image that resampling module 20 resampling module is used for segmented extraction is obtained carries out resampling；
Data extending module 30 data extending module carries out data extending to the area image of resampling；
Area image feature recognition module 40 area image feature recognition module is used for using deep learning method cog region
Area image feature.
Embodiment five
The present embodiment is to obtain in example IV, the present embodiment be a kind of based on the eye fundus image of deep learning at
Reason system, as shown in figure 4, including eye fundus image input equipment 1, eye fundus image input equipment 1 is with eye fundus image processing meanss 2 even
Connect, eye fundus image processing meanss 2 are connected with result output device 3；
Eye fundus image input equipment 1 in the present embodiment is used for obtaining eye fundus image, specific eye fundus image input equipment 1
Can be fundus camera, (abbreviation of Picture Archiving and Communication Systems, is shadow to PACS
As archiving and communication system) or image storage systems etc.；After eye fundus image input equipment 1 obtains image, by image transmitting to eye
Bottom image processing apparatus 2, wherein eye fundus image input equipment 1 pass through DICOM or TCP/IP interface and eye fundus image processing meanss 2
Connect, image can be uploaded in the way of being uploaded using automatic or manual, user also can be by Internet-browser with HTTP
Mode uploads eye fundus image；
Eye fundus image processing meanss 2 according to the eye fundus image analysis method in embodiment in the application one to embodiment three,
The eye fundus image that eye fundus image input equipment 1 transmission is come is analyzed, and analysis result is transferred to result output device 3,
Analysis result is exported by result output device 3, and specific analysis result, with forms such as image and/or statistical graphs, is passed through
The form output of DICOM browser and/or webpage.
Embodiment six
The present embodiment provides a kind of method of eye fundus image pretreatment in the application, carries out region to eye fundus image
Segmented extraction a kind of method, but in the application eye fundus image with process method can also using of the prior art other
Method.
As shown in figure 5, being the flow chart of the eye fundus image preprocess method of the present embodiment, the eye fundus image of this embodiment is pre-
Processing method comprises the following steps S 10 to step S 60：
S 10：Optical fundus blood vessel on eye fundus image for the identification；
S 20：The overall morphosiss of optical fundus blood vessel are indicated；Optical fundus blood vessel trend is indicated, obtains blood vessel
Direction collection of illustrative plates；
S 30：According to the overall morphosiss of the optical fundus blood vessel in step S 20, coarse localization optic disc area；
S 40：In the optic disc area of the coarse localization in step S 30, determine that local high-contrast or high-brightness region are
The pixel cluster of optic disc；
S 50：Determine optic disc border with ellipse fitting method on the pixel cluster of the optic disc obtaining in step S 40,
Obtain optic disc area；
S 60：Determine the elemental range of macular area in conjunction with the vessel directions collection of illustrative plates in step S 20, calculate the distance of blood vessel
, find the maximum value position of blood vessel distance field, the i.e. central recessed position of macular area in the elemental range of macular area.
Preferably, the method for optical fundus blood vessel on eye fundus image for the identification in step slo can adopt patent application
Number carry out the identification of eye fundus image blood vessel for the method recorded in 201410220540.0.Comprise the following steps：From rgb format
Retinal fundus images in extract retina gray-scale maps in green path；Multiple contrast yardsticks are set, contrast chi at each
Degree is lower all to carry out to the pixel described retina gray-scale maps contrasting metrization from multiple directions, obtains retina binaryzation pair
Ratio degree figure, the pixel of wherein said retina binarized contrast's degree in figure is divided into pixel and non-retinal blood on retinal vessel
Pixel on pipe；Difference fusion is carried out to the multiple described retina binarized contrast's degree figure obtaining, and removes acquisition after fusion
Image in noise and artifact, obtain retina fusion figure；Merge in figure from described retina and determine central retroreflective regions, and
Described central authorities retroreflective regions are filled with, obtain retinal vessel scattergram.Other blood vessels know method for distinguishing in the application
In also applicable, will not be described here.
Preferably, the overall morphosiss of the optical fundus blood vessel in step S20 are indicated method and are：Using multi-level
The form that parabola is similar to umbrella to optical fundus blood vessel is described, and as shown in Fig. 6-A, is the eye of the eye fundus image in the present embodiment
The parabola aspect graph of bottom blood vessel.S 30：According to the overall morphosiss of the optical fundus blood vessel in step S 20, coarse localization optic disc
Area, as shown in Fig. 6-A, wherein parabola vertex position corresponds to the position in optic disc area substantially；In step S20 to optical fundus blood vessel
Trend is indicated method and is：A series of discrete for the blood vessel direction turning to straightways, representing blood vessel using straightway, pass through
Count the trend that these directions determine blood vessel, that is, obtain vessel directions collection of illustrative plates；As shown in figure 6-b, it is optical fundus in the present embodiment
The direction collection of illustrative plates of the optical fundus blood vessel of image, the highlight regions in collection of illustrative plates divide 2 optic disc areas and highlight regions to divide 1 for macular area；
Preferably, the method that optical fundus blood vessel trend is indicated in step S20 is more specifically a kind of operation side
Formula：As shown in fig. 7,
Eye fundus image shown in given first Fig. 7-A, (may also be employed additive method in practice using lines approximating method
Identification blood vessel) automatic identification retinal vessel is as shown in Fig. 7-B；
Then with regular grid, blood vessel is carried out as shown in Fig. 7-B with segmentation (can be using additive method to segmentation of blood vessels), often
In individual grid, vessel segment axis direction is defined as the line direction in two solstics；.
In given image, 1 point of P (x, y), uses equation below
Calculate the orientation of vessel segment, wherein Si is the orientation of the vessel segment (Vi) of given area around set point P, Ti is point
Line between the center of P Peripheral blood pipeline section and point P.Calculate vessel segment orientation in all grids with the method, obtain as Fig. 7-C
Shown direction collection of illustrative plates；
In order to obtain metastable optic disc area, the former direction collection of illustrative plates (Fig. 7-C) in units of pixel position value is changed
Become the new direction collection of illustrative plates in units of orientation values maximum around pixel.New direction collection of illustrative plates after smooth is as shown in Fig. 7-D；
Can be seen that macular area and optic disc area have of a relatively high orientation values from the collection of illustrative plates of direction, by calculate macular area and
Distinguishing this two regions, the region 31 more than blood vessel number is considered as optic disc area to the blood vessel number in optic disc area, and 32 is macular area.
Preferably, with regard to as follows for the concrete operations of the determination of the pixel cluster of optic disc in step S 40：First, from RGB
Green channel is extracted in the colored retinal fundus images of form.Secondly, given eye fundus image pixel P (x, y), using sliding window
Port filter ((may also be employed additive method calculate generic pixel brightness)) (pixel is bright to calculate the generic pixel brightness of each pixel
Angle value and the difference of window average brightness value).Eye fundus image is applied this wave filter obtain a grey-scale contrast image.Although from this
Optic disc border can intuitively be distinguished in individual contrast image, but because the discontinuous row on optic disc border leads to still be difficult to know identification
Optic disc border, therefore we are altered in steps the contrast threshold of grey-scale contrast image to obtain the pixel accurately representing optic disc
Cluster.
Preferably, with regard to as follows for the concrete operations of the determination regarding panel boundary in step S 50：For obtain in S40
The pixel cluster of optic disc, obtains circular optic disc using Hough transformation (additive method may also be employed), on the basis of border circular areas,
Determine the neighbouring pixel cluster of this border circular areas periphery, then determine optic disc border using the method for ellipse fitting, finally by neighbour
The pixel on closely oval border also classifies as optic disc region, and the final border obtaining is optic disc region.Regard in view of in different images
The brightness of disk and contrast differ greatly, and we adopt multiple dimensioned strategy, are illustrated in figure 8 the optic disc under different brightness,
Wherein Threshold is the threshold value of brightness；Progressively adjust the brightness of optic disc, then repeat whole circle detection and ellipse fitting,
Each ellipse is scored, the standard of scoring is to consider the ratio of semi-minor axis length of ellipse and the mean flow rate of ellipse, comments
Highest is divided to be optic disc area.Using same method, determined by optic disc area can position optic cup.
Preferably, with regard to as follows for the concrete operations of the determination of macular area in step S 60：After optic disc region determines,
Macular region is substantially positioned according to the direction collection of illustrative plates of blood vessel, concrete grammar is the maximum collection of illustrative plates removing optic disc area in investigation collection of illustrative plates
Value.Afterwards, calculate the distance field of blood vessel, macula lutea approximate region is found the distance field value position of maximum, this position
It is the fovea centraliss position of macular area；
More specifically：For the feature of optical fundus blood vessel, blood vessel distal generally points to macular area position, but with regard
Disk is compared, and the blood vessel of macular area is thinner, and the brightness of image of macular area is than relatively low.In order to be accurately positioned the center of macula lutea
Domain, i.e. fovea centraliss, we utilize blood vessel distance field, as shown in figure 9, wherein Fig. 9-A is the optic disc area 52 and macular area marking off
51, Fig. 9-B are the distance field of the optical fundus blood vessel of eye fundus image corresponding with Fig. 9-A, and the 54 of circle mark is optical fundus blood vessel distance
The maximum of field, the fovea centraliss position as in macular area.
The automatic identifying method in above-mentioned optic disc area is equally applicable to the automatic identification of optic cup, in such as Fig. 9-A, in optic disc area
52 comprise optic cup 53, and (optic disc comprises optic cup：Optic disc is also named regarding nipple, and optic cup is also optic papilla sunken).
Eye fundus image analysis method in this enforcement, the multimodal system of optical fundus blood vessel is (such as：Blood vessel general morphology and blood vessel
Tendency), the image of optic disc/macular area and the features such as anatomy (such as：The region both collected positioned at blood vessel) organically combine
Get up, determine optic disc area and optic cup region first using the method for ellipse fitting, on the basis of determining optic disc region, in conjunction with
Luminance difference on image of optic disc and macula lutea and distance field analysis method, finally determine macular area.The method takes full advantage of
The form of optical fundus Various Tissues and characteristics of image, and these features are combined, make use of the complementation between these forms
Relation, so that it is guaranteed that the method does not rely on certain form or feature, may finally adapt to various picture qualities and illumination
The factors such as environment are it is ensured that accurately, stably obtain macula lutea and optic disc region.
Image pre-processing method in the application is not limited to the method in the present embodiment, additive method of the prior art
Can adopt on the basis of pretreatment as the pretreatment of eye fundus image, thus eye fundus image is carried out with the segmented extraction in region
The method of deep learning is further processed.Repeat no more again.
Compared with prior art, a kind of method for processing fundus images based on deep learning of the present invention, device and
System, has reached following effect：
(1) method for processing fundus images based on deep learning in the application, apparatus and system, it is possible to achieve automatically right
Eye fundus image is analyzed, and analysis result is objective and accurate.
(2) method for processing fundus images based on deep learning in the application, apparatus and system, using deep learning
Method, automatically study view data abstract characteristics at all levels, thus the substitutive characteristics of preferably response data, reduce people
The incompleteness causing for design feature.
(3) method for processing fundus images based on deep learning in the application, apparatus and system, in Image semantic classification mistake
Journey adds optic disc, optic cup, circumpapillary atrophy region, the segmented extraction of angiosomeses, as the data of subsequent deep learning.
So neither need downscaled images, also can complete the training of data in the acceptable time, solve deep learning in time
The problem of inadmissibility.
Because method part has been described in detail to the embodiment of the present application, here to the system being related in embodiment
Expansion description with method corresponding part is omitted, and repeats no more.Description for particular content in system refers to method enforcement
The content of example, no longer specifically limits here.
Described above illustrate and describes some preferred embodiments of the application, but as previously mentioned it should be understood that the application
Be not limited to form disclosed herein, be not to be taken as the exclusion to other embodiment, and can be used for various other combinations,
Modification and environment, and can be in application contemplated scope described herein, by technology or the knowledge of above-mentioned teaching or association area
It is modified.And the change that those skilled in the art are carried out and change without departing from spirit and scope, then all should be in this Shen
Please be in the protection domain of claims.
Claims (9)
1. a kind of method for processing fundus images based on deep learning is it is characterised in that comprise the following steps：
Eye fundus image pretreatment first, carries out the segmented extraction in region to eye fundus image；
Then area image segmented extraction being obtained carries out resampling；
Data extending is carried out to the area image of resampling；
Using deep learning method identification region characteristics of image.
2. the method for processing fundus images based on deep learning according to claim 1 is it is characterised in that described optical fundus figure
As pretreatment also includes calculating whole eye fundus image field color meansigma methodss, any pixel of eye fundus image deducts this meansigma methodss face
Color.
3. the method for processing fundus images based on deep learning according to claim 2 is it is characterised in that described region is wrapped
Include optic disc region, optic cup region, circumpapillary atrophy region and angiosomeses.
4. the method for processing fundus images based on deep learning according to claim 1 is it is characterised in that described resampling
Generate fettucelle and cut out areas image including by area image.
5. the method for processing fundus images based on deep learning according to claim 1 is it is characterised in that described data expands
Fill the regulation including area image scaling, area image rotation, area image inclination and/or areal image contrast.
6. the method for processing fundus images based on deep learning according to claim 1 is it is characterised in that described adopt deeply
Degree learning method identification region characteristics of image includes area image features training and the identification of area image feature.
7. the method for processing fundus images based on deep learning according to claim 6 is it is characterised in that described depth
Learning method is convolutional neural networks method.
8. a kind of eye fundus image processing meanss based on deep learning are it is characterised in that include：
Eye fundus image pretreatment module, described for pretreatment is carried out to eye fundus image, eye fundus image is carried out with the segmentation in region
Extract；
Resampling module, the area image that described resampling module is used for segmented extraction is obtained carries out resampling；
Data extending module, described data extending module carries out data extending to the area image of resampling；And
Area image feature recognition module, described area image feature recognition module is used for using deep learning method identification region
Characteristics of image.
9. a kind of eye fundus image processing system based on deep learning is it is characterised in that including eye fundus image input equipment, described
Eye fundus image input equipment is connected with eye fundus image processing meanss described in claim 8, described eye fundus image processing meanss and knot
Fruit output device connects.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610884671.8A CN106408564B (en) | 2016-10-10 | 2016-10-10 | A kind of method for processing fundus images based on deep learning, apparatus and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610884671.8A CN106408564B (en) | 2016-10-10 | 2016-10-10 | A kind of method for processing fundus images based on deep learning, apparatus and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106408564A true CN106408564A (en) | 2017-02-15 |
| CN106408564B CN106408564B (en) | 2019-04-02 |
Family
ID=59228993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610884671.8A Active CN106408564B (en) | 2016-10-10 | 2016-10-10 | A kind of method for processing fundus images based on deep learning, apparatus and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106408564B (en) |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106971174A (en) * | 2017-04-24 | 2017-07-21 | 华南理工大学 | A kind of CNN models, CNN training methods and the vein identification method based on CNN |
| CN107045720A (en) * | 2017-05-04 | 2017-08-15 | 深圳硅基智能科技有限公司 | Artificial neural network and system for recognizing eye fundus image lesion |
| CN107527069A (en) * | 2017-08-22 | 2017-12-29 | 京东方科技集团股份有限公司 | Image processing method, device, electronic equipment and computer-readable medium |
| CN107644418A (en) * | 2017-09-26 | 2018-01-30 | 山东大学 | Optic disk detection method and system based on convolutional neural networks |
| CN108230341A (en) * | 2018-03-07 | 2018-06-29 | 汕头大学 | A kind of eye fundus image blood vessel segmentation method that nomography is scratched based on layering |
| CN108230322A (en) * | 2018-01-28 | 2018-06-29 | 浙江大学 | A kind of eyeground feature detection device based on weak sample labeling |
| CN108492285A (en) * | 2018-03-13 | 2018-09-04 | 北京理工大学 | A kind of radial line optic disk atrophy arc dividing method based on grey scale change |
| CN108596338A (en) * | 2018-05-09 | 2018-09-28 | 四川斐讯信息技术有限公司 | A kind of acquisition methods and its system of neural metwork training collection |
| CN108717696A (en) * | 2018-05-16 | 2018-10-30 | 上海鹰瞳医疗科技有限公司 | Macula lutea image detection method and equipment |
| WO2018201633A1 (en) * | 2017-05-04 | 2018-11-08 | 深圳硅基仿生科技有限公司 | Fundus image-based diabetic retinopathy identification system |
| CN109166117A (en) * | 2018-08-31 | 2019-01-08 | 福州依影健康科技有限公司 | A kind of eye fundus image automatically analyzes comparison method and a kind of storage equipment |
| CN109325942A (en) * | 2018-09-07 | 2019-02-12 | 电子科技大学 | Eye fundus image Structural Techniques based on full convolutional neural networks |
| CN109427052A (en) * | 2017-08-29 | 2019-03-05 | 中国移动通信有限公司研究院 | Correlation technique and equipment based on deep learning processing eye fundus image |
| CN109598733A (en) * | 2017-12-31 | 2019-04-09 | 南京航空航天大学 | Retinal fundus images dividing method based on the full convolutional neural networks of depth |
| CN109685770A (en) * | 2018-12-05 | 2019-04-26 | 合肥奥比斯科技有限公司 | Retinal vessel curvature determines method |
| CN109977905A (en) * | 2019-04-04 | 2019-07-05 | 北京百度网讯科技有限公司 | Method and apparatus for handling eye fundus image |
| CN110101361A (en) * | 2019-04-23 | 2019-08-09 | 深圳市新产业眼科新技术有限公司 | Based on big data on-line intelligence diagnostic platform and its operation method and storage medium |
| CN110327013A (en) * | 2019-05-21 | 2019-10-15 | 北京至真互联网技术有限公司 | Eye fundus image detection method, device and equipment and storage medium |
| CN110400288A (en) * | 2019-06-18 | 2019-11-01 | 中南民族大学 | A kind of sugar of fusion eyes feature nets sick recognition methods and device |
| CN110399929A (en) * | 2017-11-01 | 2019-11-01 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN110489577A (en) * | 2019-08-06 | 2019-11-22 | 腾讯医疗健康(深圳)有限公司 | Medical imaging management method and device, ophthalmoscopic image processing method, electronic equipment |
| CN110543921A (en) * | 2019-10-14 | 2019-12-06 | 四川大学 | cable early fault identification method based on waveform learning |
| CN110838116A (en) * | 2019-11-14 | 2020-02-25 | 上海联影医疗科技有限公司 | Medical image acquisition method, device, equipment and computer-readable storage medium |
| CN112017168A (en) * | 2020-08-25 | 2020-12-01 | 北京致远慧图科技有限公司 | Method and device for detecting optic nerve fiber layer defect |
| CN113012093A (en) * | 2019-12-04 | 2021-06-22 | 深圳硅基智能科技有限公司 | Training method and training system for glaucoma image feature extraction |
| WO2022088856A1 (en) * | 2020-10-28 | 2022-05-05 | 北京至真互联网技术有限公司 | Fundus image recognition method and apparatus, and device |
| WO2023240674A1 (en) * | 2022-06-15 | 2023-12-21 | 潍坊眼科医院有限责任公司 | Fundus image quality optimization method based on deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1101249A (en) * | 1994-06-24 | 1995-04-12 | 中国科学院上海技术物理研究所 | Real time collecting for eyeground picture and processing method and its apparatus |
| US20050089207A1 (en) * | 2003-10-24 | 2005-04-28 | Kyoji Sekiguchi | Funduscopic image processing unit and method |
| CN104636701A (en) * | 2014-12-12 | 2015-05-20 | 浙江工业大学 | Laser two-dimension code identification method based on image restoration |
| CN105069413A (en) * | 2015-07-27 | 2015-11-18 | 电子科技大学 | Human body gesture identification method based on depth convolution neural network |
-
2016
- 2016-10-10 CN CN201610884671.8A patent/CN106408564B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1101249A (en) * | 1994-06-24 | 1995-04-12 | 中国科学院上海技术物理研究所 | Real time collecting for eyeground picture and processing method and its apparatus |
| US20050089207A1 (en) * | 2003-10-24 | 2005-04-28 | Kyoji Sekiguchi | Funduscopic image processing unit and method |
| CN104636701A (en) * | 2014-12-12 | 2015-05-20 | 浙江工业大学 | Laser two-dimension code identification method based on image restoration |
| CN105069413A (en) * | 2015-07-27 | 2015-11-18 | 电子科技大学 | Human body gesture identification method based on depth convolution neural network |
Non-Patent Citations (1)
| Title |
|---|
| 高玮玮: "眼底图像分割算法及应用研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (54)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106971174B (en) * | 2017-04-24 | 2020-05-22 | 华南理工大学 | CNN model, CNN training method and CNN-based vein identification method |
| CN106971174A (en) * | 2017-04-24 | 2017-07-21 | 华南理工大学 | A kind of CNN models, CNN training methods and the vein identification method based on CNN |
| CN107358606B (en) * | 2017-05-04 | 2018-07-27 | 深圳硅基仿生科技有限公司 | The artificial neural network device and system and device of diabetic retinopathy for identification |
| CN107045720A (en) * | 2017-05-04 | 2017-08-15 | 深圳硅基智能科技有限公司 | Artificial neural network and system for recognizing eye fundus image lesion |
| CN107045720B (en) * | 2017-05-04 | 2018-11-30 | 深圳硅基仿生科技有限公司 | The processing system of identification eye fundus image lesion based on artificial neural network |
| WO2018201633A1 (en) * | 2017-05-04 | 2018-11-08 | 深圳硅基仿生科技有限公司 | Fundus image-based diabetic retinopathy identification system |
| WO2018201632A1 (en) * | 2017-05-04 | 2018-11-08 | 深圳硅基仿生科技有限公司 | Artificial neural network and system for recognizing lesion in fundus image |
| CN107358606A (en) * | 2017-05-04 | 2017-11-17 | 深圳硅基智能科技有限公司 | For identifying the artificial neural network and system of diabetic retinopathy |
| CN107358605B (en) * | 2017-05-04 | 2018-09-21 | 深圳硅基仿生科技有限公司 | The deep neural network apparatus and system of diabetic retinopathy for identification |
| CN107358605A (en) * | 2017-05-04 | 2017-11-17 | 深圳硅基智能科技有限公司 | For identifying the deep neural network and system of diabetic retinopathy |
| US11636664B2 (en) | 2017-08-22 | 2023-04-25 | Boe Technology Group Co., Ltd. | Training method and device of neural network for medical image processing, and medical image processing method and device |
| CN107527069A (en) * | 2017-08-22 | 2017-12-29 | 京东方科技集团股份有限公司 | Image processing method, device, electronic equipment and computer-readable medium |
| CN109427052A (en) * | 2017-08-29 | 2019-03-05 | 中国移动通信有限公司研究院 | Correlation technique and equipment based on deep learning processing eye fundus image |
| CN109427052B (en) * | 2017-08-29 | 2020-10-02 | 中国移动通信有限公司研究院 | Related method and device for processing fundus images based on deep learning |
| CN107644418B (en) * | 2017-09-26 | 2019-11-08 | 山东大学 | Optic disk detection method and system based on convolutional neural networks |
| CN107644418A (en) * | 2017-09-26 | 2018-01-30 | 山东大学 | Optic disk detection method and system based on convolutional neural networks |
| CN110399929A (en) * | 2017-11-01 | 2019-11-01 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN109598733A (en) * | 2017-12-31 | 2019-04-09 | 南京航空航天大学 | Retinal fundus images dividing method based on the full convolutional neural networks of depth |
| CN108230322B (en) * | 2018-01-28 | 2021-11-09 | 浙江大学 | Eye ground characteristic detection device based on weak sample mark |
| CN108230322A (en) * | 2018-01-28 | 2018-06-29 | 浙江大学 | A kind of eyeground feature detection device based on weak sample labeling |
| CN108230341A (en) * | 2018-03-07 | 2018-06-29 | 汕头大学 | A kind of eye fundus image blood vessel segmentation method that nomography is scratched based on layering |
| CN108230341B (en) * | 2018-03-07 | 2021-12-17 | 汕头大学 | Eyeground image blood vessel segmentation method based on layered matting algorithm |
| CN108492285A (en) * | 2018-03-13 | 2018-09-04 | 北京理工大学 | A kind of radial line optic disk atrophy arc dividing method based on grey scale change |
| CN108492285B (en) * | 2018-03-13 | 2021-11-09 | 北京理工大学 | Radial optic disc atrophy arc segmentation method based on gray scale change |
| CN108596338A (en) * | 2018-05-09 | 2018-09-28 | 四川斐讯信息技术有限公司 | A kind of acquisition methods and its system of neural metwork training collection |
| CN108717696B (en) * | 2018-05-16 | 2022-04-22 | 上海鹰瞳医疗科技有限公司 | Yellow spot image detection method and equipment |
| CN108717696A (en) * | 2018-05-16 | 2018-10-30 | 上海鹰瞳医疗科技有限公司 | Macula lutea image detection method and equipment |
| CN109166117B (en) * | 2018-08-31 | 2022-04-12 | 福州依影健康科技有限公司 | Automatic eye fundus image analysis and comparison method and storage device |
| CN109166117A (en) * | 2018-08-31 | 2019-01-08 | 福州依影健康科技有限公司 | A kind of eye fundus image automatically analyzes comparison method and a kind of storage equipment |
| CN109325942A (en) * | 2018-09-07 | 2019-02-12 | 电子科技大学 | Eye fundus image Structural Techniques based on full convolutional neural networks |
| CN109325942B (en) * | 2018-09-07 | 2022-03-25 | 电子科技大学 | Fundus image structure segmentation method based on full convolution neural network |
| CN109685770A (en) * | 2018-12-05 | 2019-04-26 | 合肥奥比斯科技有限公司 | Retinal vessel curvature determines method |
| CN109685770B (en) * | 2018-12-05 | 2020-10-09 | 合肥奥比斯科技有限公司 | Method for determining retinal vascular tortuosity |
| CN109977905B (en) * | 2019-04-04 | 2021-08-06 | 北京百度网讯科技有限公司 | Method and apparatus for processing fundus images |
| CN109977905A (en) * | 2019-04-04 | 2019-07-05 | 北京百度网讯科技有限公司 | Method and apparatus for handling eye fundus image |
| US11232560B2 (en) | 2019-04-04 | 2022-01-25 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method and apparatus for processing fundus image |
| CN110101361B (en) * | 2019-04-23 | 2022-07-12 | 深圳市新产业眼科新技术有限公司 | Big data based online intelligent diagnosis platform and operation method and storage medium thereof |
| CN110101361A (en) * | 2019-04-23 | 2019-08-09 | 深圳市新产业眼科新技术有限公司 | Based on big data on-line intelligence diagnostic platform and its operation method and storage medium |
| CN110327013B (en) * | 2019-05-21 | 2022-02-15 | 北京至真互联网技术有限公司 | Fundus image detection method, device and equipment and storage medium |
| CN110327013A (en) * | 2019-05-21 | 2019-10-15 | 北京至真互联网技术有限公司 | Eye fundus image detection method, device and equipment and storage medium |
| CN110400288B (en) * | 2019-06-18 | 2021-08-10 | 中南民族大学 | Sugar network disease identification method and device fusing binocular features |
| CN110400288A (en) * | 2019-06-18 | 2019-11-01 | 中南民族大学 | A kind of sugar of fusion eyes feature nets sick recognition methods and device |
| CN110489577B (en) * | 2019-08-06 | 2024-01-26 | 腾讯医疗健康(深圳)有限公司 | Medical image management method and device, fundus image processing method and electronic equipment |
| CN110489577A (en) * | 2019-08-06 | 2019-11-22 | 腾讯医疗健康(深圳)有限公司 | Medical imaging management method and device, ophthalmoscopic image processing method, electronic equipment |
| CN110543921A (en) * | 2019-10-14 | 2019-12-06 | 四川大学 | cable early fault identification method based on waveform learning |
| CN110838116A (en) * | 2019-11-14 | 2020-02-25 | 上海联影医疗科技有限公司 | Medical image acquisition method, device, equipment and computer-readable storage medium |
| CN110838116B (en) * | 2019-11-14 | 2023-01-03 | 上海联影医疗科技股份有限公司 | Medical image acquisition method, device, equipment and computer-readable storage medium |
| CN113012093A (en) * | 2019-12-04 | 2021-06-22 | 深圳硅基智能科技有限公司 | Training method and training system for glaucoma image feature extraction |
| CN113011450A (en) * | 2019-12-04 | 2021-06-22 | 深圳硅基智能科技有限公司 | Training method, training device, recognition method and recognition system for glaucoma recognition |
| CN113012093B (en) * | 2019-12-04 | 2023-12-12 | 深圳硅基智能科技有限公司 | Training method and training system for glaucoma image feature extraction |
| CN112017168A (en) * | 2020-08-25 | 2020-12-01 | 北京致远慧图科技有限公司 | Method and device for detecting optic nerve fiber layer defect |
| WO2022088856A1 (en) * | 2020-10-28 | 2022-05-05 | 北京至真互联网技术有限公司 | Fundus image recognition method and apparatus, and device |
| US11620763B2 (en) | 2020-10-28 | 2023-04-04 | Beijing Zhenhealth Technology Co., Ltd. | Method and device for recognizing fundus image, and equipment |
| WO2023240674A1 (en) * | 2022-06-15 | 2023-12-21 | 潍坊眼科医院有限责任公司 | Fundus image quality optimization method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106408564B (en) | 2019-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106408564B (en) | A kind of method for processing fundus images based on deep learning, apparatus and system | |
| CN106920227B (en) | The Segmentation Method of Retinal Blood Vessels combined based on deep learning with conventional method | |
| WO2020259209A1 (en) | Fundus image recognition method, apparatus and device, and storage medium | |
| CN107886503A (en) | A kind of alimentary canal anatomical position recognition methods and device | |
| CN108095683A (en) | The method and apparatus of processing eye fundus image based on deep learning | |
| CN109166124A (en) | A kind of retinal vascular morphologies quantization method based on connected region | |
| CN109635862A (en) | Retinopathy of prematurity plus lesion classification method | |
| CN107330876A (en) | A kind of image automatic diagnosis method based on convolutional neural networks | |
| Sertkaya et al. | Diagnosis of eye retinal diseases based on convolutional neural networks using optical coherence images | |
| CN108053398A (en) | A kind of melanoma automatic testing method of semi-supervised feature learning | |
| CN109919938B (en) | Method for obtaining optic disc segmentation atlas of glaucoma | |
| CN109583331B (en) | Deep learning-based accurate positioning method for positions of wrist vein and mouth of person | |
| CN108765392A (en) | A kind of digestive endoscope lesion detection and recognition methods based on sliding window | |
| CN107292835A (en) | A kind of method and device of eye fundus image retinal vessel Automatic Vector | |
| CN102567734A (en) | Specific value based retina thin blood vessel segmentation method | |
| CN106846293A (en) | Image processing method and device | |
| CN106780439A (en) | A kind of method for screening eye fundus image | |
| CN110827273A (en) | Tea disease detection method based on regional convolution neural network | |
| CN106355574A (en) | Intra-abdominal adipose tissue segmentation method based on deep learning | |
| CN112233087A (en) | Artificial intelligence-based ophthalmic ultrasonic disease diagnosis method and system | |
| CN106570530A (en) | Extraction method for extracting hard exudates in ophthalmoscopic image | |
| CN106960199B (en) | Complete extraction method of white eye region of true color eye picture | |
| CN117764957A (en) | Glaucoma image feature extraction training system based on artificial neural network | |
| CN111028230A (en) | Fundus image optic disc and macula lutea positioning detection algorithm based on YOLO-V3 | |
| CN106446805A (en) | Segmentation method and system for optic cup in eye ground photo |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |