CN109833025A - A kind of method for detecting abnormality of retina, device, equipment and storage medium - Google Patents
A kind of method for detecting abnormality of retina, device, equipment and storage medium Download PDFInfo
- Publication number
- CN109833025A CN109833025A CN201910251114.6A CN201910251114A CN109833025A CN 109833025 A CN109833025 A CN 109833025A CN 201910251114 A CN201910251114 A CN 201910251114A CN 109833025 A CN109833025 A CN 109833025A
- Authority
- CN
- China
- Prior art keywords
- image
- retinal
- retina
- retinal images
- former
- 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
Abstract
The embodiment of the invention discloses a kind of method for detecting abnormality of retina, device, equipment and storage mediums, which comprises obtains the former retinal images of a retina；Determine the picture quality of original retinal images；When picture quality meets preset quality requirements, the pretreatment of image Style Transfer is carried out to former retinal images, obtains targeted retinal image；The exception information of retina is determined according to targeted retinal image.It realizes before the abnormality detection of retina, it determines the picture quality figure of retinal images and Style Transfer pretreatment is carried out to retinal images, so that the retinal images for determining retinal abnormalities are high quality and the consistent image of style, solve retinal image quality difference and stylistic differences, cause the problem of the abnormality detection inaccuracy of retina, abnormality detection of enough sharp picture features for retina can be extracted from high quality and the consistent targeted retinal image of style, improve the accuracy of retinal abnormalities detection.
Description
Technical field
The present embodiments relate to technical field of image processing more particularly to a kind of method for detecting abnormality of retina, view
Abnormal detector, equipment and the storage medium of nethike embrane.
Background technique
Retinal abnormalities, such as diabetic retinopathy (Diabetic Retinopathy, DR), treating senile maculopathy
(Age-related Macular Degeneration, AMD), glaucoma, pathological myopia (Pathological
Myopia, PM) etc. be the main reason for leading to visual impairment in world wide, while one of the main reason for be also blindness.
For the exception of above-mentioned retina, whether the retina that user can be detected based on retinal images is abnormal, thus
Prevent user's eyesight and is further damaged even blinding.However, the Outlier Detection Algorithm based on retinal images is to use to disclose
Limited retinal images training, detection device hardware have differences with Image Acquisition personnel acquire image gimmick it is different
In the case of, what the retinal images of collection in worksite to user were used there are of poor quality, collected retinal images and when training
Training image problem inconsistent in image style, so that the generalization and shifting of the Outlier Detection Algorithm based on retinal images
Plant property is poor, and enough sharp picture features can not be extracted from collected retinal images for abnormality detection, to drop
The accuracy of low retinal abnormalities detection.
To sum up, since there are the differences in of poor quality and style for collected retinal images, to cause retina
The problem of abnormality detection inaccuracy.
Summary of the invention
The embodiment of the invention provides a kind of method for detecting abnormality of retina, device, equipment and storage mediums, to realize
The exception of retina is accurately detected, retinal images in the prior art is solved and is caused there are of poor quality and stylistic differences
The problem of the abnormality detection inaccuracy of retina.
In a first aspect, the embodiment of the invention provides a kind of method for detecting abnormality of retina, comprising:
Obtain the former retinal images of a retina；
Determine the picture quality of the former retinal images；
When described image quality meets preset quality requirements, image Style Transfer is carried out to the former retinal images
Pretreatment, obtains targeted retinal image；
The exception information of the retina is determined according to the targeted retinal image.
Second aspect, the embodiment of the invention provides a kind of abnormal detectors of retina, comprising:
Former retinal images obtain module, for obtaining the former retinal images of a retina；
Picture quality determining module, for determining the picture quality of the former retinal images；
Image Style Transfer module, for when described image quality meets preset quality requirements, to the former view
Film image carries out the pretreatment of image Style Transfer, obtains targeted retinal image；
Exception information determining module, for determining the exception information of the retina according to the targeted retinal image.
The third aspect, the embodiment of the invention provides a kind of equipment, the equipment includes:
One or more processors；
Storage device, for storing one or more programs；
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device realizes the method for detecting abnormality of retina described in any embodiment of that present invention.
Fourth aspect, the embodiment of the invention provides a kind of computer readable storage mediums, are stored thereon with computer journey
Sequence realizes the method for detecting abnormality of retina described in any embodiment of that present invention when the program is executed by processor.
The embodiment of the invention provides a kind of method for detecting abnormality of retina, by the former retina for obtaining a retina
Image, and determine the picture quality of former retinal images, when picture quality meets preset quality requirements, to former retinal map
As carrying out the pretreatment of image Style Transfer, targeted retinal image is obtained, and then retina is determined according to targeted retinal image
Exception information, realize before the abnormality detection of retina, determine the picture quality figure of retinal images and to retinal map
As carrying out Style Transfer pretreatment, so that the retinal images for determining retinal abnormalities are high quality and the consistent figure of style
Picture solves retinal images there are of poor quality and stylistic differences, causes the problem to the abnormality detection inaccuracy of retina, energy
It is enough that exception of enough sharp picture features for retina is extracted from high quality and the consistent targeted retinal image of style
Detection improves the accuracy of retinal abnormalities detection.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, of the invention other
Feature, objects and advantages will become more apparent upon:
Figure 1A is a kind of flow chart of the method for detecting abnormality for retina that the embodiment of the present invention one provides；
Figure 1B is the schematic diagram of retinal images provided in an embodiment of the present invention；
Fig. 1 C is that hand-held eyeground equipment acquires image to the schematic diagram for storing image in the embodiment of the present invention；
Fig. 2A is a kind of flow chart of the method for detecting abnormality of retina provided by Embodiment 2 of the present invention；
Fig. 2 B is in the embodiment of the present invention two for determining the schematic diagram of the deep neural network of image quality measure value；
Fig. 2 C is the schematic diagram that image Style Transfer model is determined in the embodiment of the present invention two；
Fig. 2 D is the schematic diagram of original retinal images and targeted retinal image in the embodiment of the present invention two；
Fig. 2 E is the schematic diagram for determining the exception information of retina in the embodiment of the present invention two based on retinal images；
Fig. 3 is a kind of structural schematic diagram of the abnormal detector for retina that the embodiment of the present invention three provides；
Fig. 4 is a kind of structural schematic diagram for equipment that the embodiment of the present invention four provides.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limiting the invention.It also should be noted that in order to just
Only the parts related to the present invention are shown in description, attached drawing rather than entire infrastructure.
Embodiment one
Figure 1A is a kind of flow chart of the method for detecting abnormality for retina that the embodiment of the present invention one provides, and the present embodiment can
Suitable for the applicable cases of abnormality detection for carrying out retina by retinal images.Retina provided in this embodiment it is different
Normal detection method can be executed by the abnormal detector of retina provided in an embodiment of the present invention, which can be by soft
The mode of part and/or hardware is realized, and is integrated in an equipment, such as handheld device, server, work station.Specifically,
With reference to Figure 1A, this method be may include steps of:
S101, the former retinal images for obtaining a retina.
Wherein, former retinal images are the figure for carrying out shooting acquisition to the left eye of user and right eye eyeground using fundus camera
Picture is as shown in Figure 1B the schematic diagram of normal retinal images and various abnormal retinal images, normal retinal map
As display includes that physiological structures, the abnormal retinal images such as optic disk, blood vessel, retina, retinal vessel, macula lutea are then shown respectively
Kind anomaly sxtructure.(a) is normal retinal images in Figure 1B, (b) is the retinal map with Diabetic retinopathy
Picture (c) is the retinal images with Age related macular disease, is (d) retinal images with glaucoma disease, (e) is
Retinal images with pathological myopia disease.
Specifically, the embodiment of the present invention can be acquired in the abnormality detecting process for carrying out retina by fundus camera
The left eye of user and the retinal images of right eye and storage.As shown in Figure 1 C, fundus camera can integrate in moveable hand-held
In the equipment of formula eyeground, after hand-held eyeground equipment acquires the retinal images of user, it on the one hand can pass through data line, wired network
The connection types such as network, wireless network are connect with cloud server, and collected retinal images are transmitted to cloud server,
The abnormality detection processing of retina is carried out on server beyond the clouds based on retinal images；On the other hand, it can also will acquire
To retinal images store into the local storage medium of hand-held eyeground equipment, to be regarded in the equipment of hand-held eyeground
The abnormality detection of nethike embrane is handled.
Certainly, fundus camera can also be integrated in desk-top eyeground equipment, and the view of user is acquired by desk-top eyeground equipment
Nethike embrane image, hand-held eyeground equipment is low relative to desk-top eyeground equipment cost, small in size, facilitates doctor outgoing medical, is applicable in
Medical hardware scarcity of resources area, and desk-top eyeground equipment is then suitable for the indoor spaces of hospital, in practical applications, Ke Yigen
According to the retinal images needed using desk-top eyeground equipment or hand-held eyeground equipment acquisition user.
S102, the picture quality for determining the former retinal images.
In embodiments of the present invention, the nerve of the image quality measure value for obtaining retinal images can be trained in advance
Former retinal images can be inputted acquisition picture quality in the neural network and commented by network after getting former retinal images
Valuation, and determine whether picture quality meets preset quality requirements according to the assessed value, meet preset matter in picture quality
When amount condition, illustrates the structure feature that former retinal images can clearly, completely show retina for abnormality detection, then hold
Otherwise row S103 returns to S101, such as generates the prompt information for reacquiring retinal images, to prompt user to resurvey use
The retinal images at family.
The embodiment of the present invention, can be deficient to avoid medical resource from far-off regions by the picture quality of determining former retinal images
Weary, personnel acquire the low-quality problem of the unprofessional retinal images for causing acquisition of image, and then avoid using low quality
Retinal images when carrying out abnormality detection, accurate characteristics of image can not be extracted from low-quality retinal images to be caused to regard
The accuracy rate of the abnormality detection of nethike embrane is low, and it is special that enough sharp pictures can be extracted from the retinal images for meet quality requirements
Sign, improves the accuracy rate of the abnormality detection of retina.
S103, when described image quality meets preset quality requirements, image wind is carried out to the former retinal images
Lattice migration pretreatment, obtains targeted retinal image.
In embodiments of the present invention, image style moves the image style that can be using algorithm learning objective image, then
The image style is applied on an other image, wherein image style can be the styles such as the color of target image, brightness,
For example color is partially warm, brightness is partially bright etc..Image Style Transfer is converted in style and keeps the content information in image
It is constant.
It is inputted specific to the former retinal images that picture quality in the embodiment of the present invention, can be met to preset quality condition
In advance in trained image composer, image Style Transfer is carried out to former retinal images by the image composer, generates mesh
Mark retinal images.Wherein, on the basis of former retinal images, the structure feature of retina is kept not targeted retinal image
Become, is converted on the images display styles such as color, brightness.
It is pre-processed by image Style Transfer, acquires gimmick when eliminating user using hand-held eyeground equipment acquisition image
Inconsistent and image capture device can be obtained because of difference of the former retinal images in style acquired caused by hardware differences
The targeted retinal image of the style of Outlier Detection Algorithm suitable for retina, so that Outlier Detection Algorithm is applicable to difference
Hardware device and different horizontal user, improve the generalization and portability of Outlier Detection Algorithm, abnormality detection is calculated
Method can extract enough sharp picture features from the former retinal images for meeting its style, further improve retina
The accuracy rate of abnormality detection.
S104, the exception information that the retina is determined according to the targeted retinal image.
In practical applications, the targeted retinal image of the left eye and right eye of available user, then regards according to target
Nethike embrane image determines the exception information of left eye and right eye, it is alternatively possible to the mould that the input of targeted retinal image is trained in advance
Exception information is extracted in type, wherein exception information may include abnormal title and abnormal value-at-risk, by left and right and right eye
Exception reporting can be generated after the exception information combination of retina.
The embodiment of the invention provides a kind of method for detecting abnormality of retina, the abnormality detection in retina is realized
Before, it determines the picture quality figure of retinal images and Style Transfer pretreatment is carried out to retinal images, so that for determining view
The retinal images of nethike embrane exception are high quality and the consistent image of style, and solving retinal images, there are of poor quality and styles
Difference causes the problem to the abnormality detection inaccuracy of retina, can be from high quality and the consistent targeted retinal figure of style
Abnormality detection of enough sharp picture features for retina is extracted as in, improves the accuracy of retinal abnormalities detection.
Embodiment two
Fig. 2A is a kind of flow chart of the method for detecting abnormality of retina provided by Embodiment 2 of the present invention, and the present embodiment exists
On the basis of embodiment one, target view is generated to the picture quality and the pretreatment of image Style Transfer that determine former retinal images
Film image optimizes, specifically, this method may include steps of with reference to Fig. 2A:
S201, the former retinal images for obtaining a retina.
S202, the former retinal images are handled to determine assessed value, the assessed value is for expressing the original
The picture quality of retinal images.
In embodiments of the present invention, a deep neural network can be trained to obtain the assessed value of picture quality in advance, had
Body, a certain number of retinal images of fundus camera acquisition, which can be used, as training image X, and according to training image X is
The no abnormality detection that can be used for retina is labeled, and obtains the label Y that each training image answers image X.By training image X and
Label Y collectively constitutes training sample, after training image X is inputted deep neural network, obtains predicted value by propagated forward
Y1, then by the penalty values of loss function calculating Y and Y1, using the penalty values backpropagation, optimize depth by successive ignition
Neural network parameter finally obtains optimal model parameters.
It is as shown in Figure 2 B the schematic diagram for obtaining the deep neural network of the assessed value of picture quality, in fig. 2b,
Former retinal images can be inputted in the deep neural network, the different layers convolution kernel in deep neural network carries out convolution behaviour
Make, successively obtain the feature vector of the different layers of former retinal images, and feature vector is input in full articulamentum, this connects entirely
The nodal point number for connecing layer input is the number of feature vector, and output node number is 2, represents 2 kinds of image quality measure values.
S203, judge whether the assessed value is preset target value.
In one example, to save Computing resource, the assessed value for expressing picture quality can be set to 0
Or 1, wherein if assessed value is 1, the quality of retinal images is better, can be used for the abnormality detection of retina, executes S204；
If assessed value is 0, the second-rate of retinal images is represented, is not useable for the abnormality detection of retina, executes S205.
Certainly, other than 0 and 1, other assessed values can also be set, for example, representing excellent 0, representing good 1, represent
2, etc. of difference, the embodiments of the present invention are not limited thereto.
S204, determine that described image quality meets preset quality requirements.
If the assessed value of picture quality is target value, such as assessed value is 1 in embodiments of the present invention, then representative image
Quality meets preset quality requirements, can be used for the abnormality detection of retina, executes S206.
S205, it determines that institute's commentary picture quality does not meet preset quality requirements, returns to S101.
If the assessed value of picture quality is not target value, such as assessed value is 0 in embodiments of the present invention, then represents figure
Image quality amount meets not preset quality requirements, the retinal images for needing that user is prompted to resurvey user, to be met
The retinal images of quality requirements then can be generated the prompt information for resurveying retinal images and return to S201.
In the embodiment of the present invention, by carrying out the assessed value that processing obtains picture quality to former retinal images, assessing
It determines that the picture quality of former retina meets preset quality condition when value is goal-based assessment value, can be used for the abnormal inspection of retina
It surveys, otherwise picture quality does not meet preset quality condition, then returns to the step of reacquiring retinal images, can be to avoid use
Low-quality retinal images carry out abnormality detection the problem for causing accuracy rate low, improve the accurate of the abnormality detection of retina
Rate.
S206, determine that image Style Transfer model, described image Style Transfer model are used for image Style Transfer.
In embodiments of the present invention, image Style Transfer model can carry out image style to the image being input in model
Migration process, so that the image after Style Transfer and image used in the Outlier Detection Algorithm of retina are consistent in style.
Specifically, as shown in Figure 2 C, available first training retinal images X, and determine the first image composer GXY
With the second image composer GYX。
Wherein, the first image composer GXYIt is pre-processed for image Style Transfer, the second image composer GYXFor image
Restore style.
In one example, the first image composer GXYWith the second image composer GYXIt can be deep neural network,
It can be machine learning model, such as SVM (Support Vector Machine, support vector machines), a kind of adaboost (iteration
Algorithm) etc., the embodiments of the present invention are not limited thereto.
First training retinal images X is inputted into the first image composer GXYMiddle progress Style Transfer pretreatment, is schemed
As the second training retinal images Y after Style Transfer, the second training retinal images Y is inputted into the second image composer GYXIn
The reduction of image style is carried out, third training retinal images X ' is obtained, then using the first training retinal images X, the second instruction
Practice retinal images Y and third training retinal images X ' to the first image composer GXYIt is adjusted, after final adjustment
First image composer GXYAs image Style Transfer model.
To the first image composer GXYIt, can be by models such as network VGG19 trained in advance from the during adjustment
The first retinal structure feature is extracted in one training retinal images X, and extracts second from the second training retinal images Y
Retinal structure feature, and the first-loss value between the first retinal structure feature and the second retinal structure feature is calculated,
The second penalty values between the first training retinal images X and third training retinal images X ' are calculated simultaneously, then using the
One penalty values and the second penalty values are to the first image composer GXYIt is adjusted, for example, being lost using first-loss value and second
Value is to the first image composer GXYBackpropagation is carried out, the first image composer G is optimizedXYParameter, obtain the first optimal figure
As generator GXYAs image Style Transfer model.
In practical applications, image Style Transfer model can be also possible to on-line training, be worked as off-line training with off-line training
After good image Style Transfer model, the image Style Transfer model can be called directly.
S207, it will be handled in the former retinal images input described image Style Transfer model, to generate target
Retinal images.
As shown in Figure 2 D, by trained image Style Transfer model, original retinal map shown in (a) in Fig. 2 D is inputted
Targeted retinal image shown in (b) in Fig. 2 D then can be generated in picture, and former retinal images and targeted retinal image are in structure
On be consistent, migrated in image style, for example, in Fig. 2 D original retinal images shown in (a) image style
Partially dark, targeted retinal image shown in (b) is bright relative to former retinal images in Fig. 2 D after image Style Transfer.
In the embodiment of the present invention, the pretreatment of image Style Transfer can be generated the image Style Transfer of former retinal images
Targeted retinal image, the style of image used in the image style and retinal abnormalities detection algorithm of targeted retinal image
Unanimously, so that retinal abnormalities detection algorithm has good generalization and transplantability, acquisition retinal images are avoided
Equipment there are the differences on hardware to cause retinal images style inconsistent, cause retinal abnormalities detection algorithm performance decline
The low problem of preparation rate can obtain the targeted retinal image of suitable style, Ke Yicong by the pretreatment of image Style Transfer
Enough, accurate characteristics of image is extracted in the targeted retinal image, improves the accuracy rate of retinal abnormalities detection.
S208, determine whether the retina is abnormal based on the targeted retinal image.
In the embodiment of the present invention, after obtaining targeted retinal image, retinal images can be inputted trained in advance
In deep neural network and classifier, determine whether retina is abnormal by deep neural network and classifier, if it is determined that view
Nethike embrane then executes S209 extremely.For example, targeted retinal image is inputted in deep neural network and classifier, retina is predicted
Abnormal value-at-risk, determines retinal abnormalities if value-at-risk is greater than preset threshold value, otherwise determines that retina is normal.
S209, it the targeted retinal image is inputted respectively in multiple preset predicting abnormality networks handles, obtain
To multiple points of feature vectors, described point of feature vector is in a kind of abnormal feature for expressing the retina.
In practical applications, retinal abnormalities may include a variety of, then can input targeted retinal image respectively and be
In a variety of extremely trained predicting abnormality networks, with extracted from targeted retinal image belong to each abnormal dtex levy to
Amount, for example, inputting DR, AMD, glaucoma and PM predicting abnormality network respectively extracts relevant point of feature vector.
S210, dtex sign Vector Groups are combined into total characteristic vector.
What multiple predicting abnormality networks extracted is to belong to corresponding point of feature vector of every kind of exception, then can be by each dtex
Sign Vector Groups are combined into total characteristic vector.
S211, it will handle, obtain more in the preset full Connection Neural Network classifier of total characteristic vector input
A value-at-risk, the value-at-risk are in a kind of abnormal risk for expressing the retina.
After obtaining total characteristic vector, a total characteristic vector can be added to input in full Connection Neural Network classifier, to obtain
The value-at-risk for obtaining various retinal abnormalities improves view without individually predicting the value-at-risk of each retinal abnormalities
The efficiency of nethike embrane abnormality detection.
In order to which those skilled in the art are more clearly understood that S208-S211, below in conjunction with Fig. 2 C to S208-S211 into
Row illustrates.
As shown in Figure 2 E, after obtaining targeted retinal image Y, by targeted retinal image input deep neural network and
The value-at-risk of retinal abnormalities is extracted in classifier, if the value-at-risk is greater than preset threshold value, for example value-at-risk is greater than 40%,
It is abnormal then to determine that the retina exists, then targeted retinal image Y is inputted into DR predicting abnormality network, AMD predicting abnormality respectively
In network, glaucoma predicting abnormality network and PM predicting abnormality network, extracts divide feature vector respectively, a point feature vector is connected
It inputs in classifier afterwards and carries out classification prediction, it is abnormal in DR exception, AMD exception, glaucoma exception and PM to respectively obtain retina
Value-at-risk above-mentioned processing is executed to the targeted retinal image of the left eye of user and right eye respectively for a user
To obtain the exception information of the retina of user's left eye and right eye, for example, left eye and the corresponding DR exception of right eye, AMD are different
Often, the value-at-risk of glaucoma exception and PM exception.
In the embodiment of the present invention, first determine whether retina is abnormal based on targeted retinal image, in retinal abnormalities
Corresponding point of feature vector of a variety of exceptions that retina is extracted by multiple preset predicting abnormality networks, will divide feature vector
Group is combined into value-at-risk of the input classifier prediction retina in a variety of exceptions after total characteristic vector, on the one hand, first determines view
Whether film is abnormal to determine each abnormal value-at-risk again, the demand of doctor is more in line with, on the other hand, once to a variety of abnormal risks
Value is predicted, the efficiency of retinal abnormalities detection can be improved, and doctor can also disposably know a variety of abnormal value-at-risks,
Exception is repeatedly determined without doctor, alleviates the workload of doctor.
Embodiment three
Fig. 3 is a kind of structural schematic diagram of the abnormal detector for retina that the embodiment of the present invention three provides, specifically,
As shown in figure 3, the apparatus may include:
Former retinal images obtain module 301, for obtaining the former retinal images of a retina；
Picture quality determining module 302, for determining the picture quality of the former retinal images；
Image Style Transfer module 303, for being regarded to the original when described image quality meets preset quality requirements
Nethike embrane image carries out the pretreatment of image Style Transfer, obtains targeted retinal image；
Exception information determining module 304, for determining that the abnormal of the retina is believed according to the targeted retinal image
Breath.
Optionally, picture quality determining module 302 includes:
Assessed value determines submodule, for being handled the former retinal images to determine assessed value, the assessment
Value is for expressing the picture quality of the former retinal images；
Judging submodule, for judging whether the assessed value is preset target value；
First picture quality determines submodule, for determining that described image quality meets preset quality requirements；
Second picture quality determines submodule, and for determining, commentary picture quality does not meet preset quality requirements, returns
It returns former retinal images and obtains module.
Optionally, described image Style Transfer module 303 includes:
Style Transfer model determines submodule, for determining image Style Transfer model, described image Style Transfer model
It is pre-processed for image Style Transfer；
Targeted retinal image generates submodule, for the former retinal images to be inputted described image Style Transfer mould
It is handled in type, to generate targeted retinal image.
Optionally, the Style Transfer model determines that submodule includes:
First training image acquiring unit, for obtaining the first training retinal images；
Image composer determination unit, for determining the first image composer, the second image composer, the first image
Generator is pre-processed for image Style Transfer, and second image composer is used for image restoring style；
Second training image acquiring unit, for generating the first training retinal images input the first image
Style Transfer is carried out in device, obtains the second training retinal images；
Third training image acquiring unit is generated for the second training retinal images to be inputted second image
Style reduction is carried out in device, obtains third training retinal images；
Adjustment unit, for using the first training retinal images, the second training retinal images and institute
It states third training retinal images to be adjusted the first image generator, to obtain image Style Transfer model.
Optionally, institute's adjustment unit includes:
Structure feature extracts subelement, for extracting the first retinal structure spy from the first training retinal images
Sign, and the second retinal structure feature is extracted from the second training retinal images；
First-loss value computation subunit, for calculating the first retinal structure feature and the second retina knot
First-loss value between structure feature；
Second penalty values computation subunit, for calculating the first training retinal images and third training view
The second penalty values between film image；
Subelement is adjusted, for using the first-loss value and second penalty values to the first image generator
It is adjusted, to obtain image Style Transfer model.
Optionally, further includes:
Value-at-risk determining module, for determining the value-at-risk of the retinal abnormalities based on the targeted retinal image,
Value-at-risk judgment module enters exception information and determines mould if being greater than preset threshold value for the value-at-risk
Block.
Optionally, the exception information determining module 304 includes:
Divide feature vector extracting sub-module, it is pre- for the targeted retinal image to be inputted multiple preset exceptions respectively
It is handled in survey grid network, obtains multiple points of feature vectors, described point of feature vector is in one kind for expressing the retina
Abnormal feature；
Total characteristic Vector Groups zygote module, for dtex sign Vector Groups to be combined into total characteristic vector；
Value-at-risk acquisition submodule, for inputting the total characteristic vector in preset full Connection Neural Network classifier
It is handled, obtains multiple value-at-risks, the value-at-risk is in a kind of abnormal risk for expressing the retina.
The retina that above-mentioned any embodiment provides can be performed in the abnormal detector of retina provided in this embodiment
Method for detecting abnormality has corresponding function and beneficial effect.
Example IV
Referring to Fig. 4, the structural schematic diagram of one of an example of the present invention equipment is shown.As shown in figure 4, the equipment
Can specifically include: processor 40, memory 41, the display screen 42 with touch function, input unit 43, output device 44 with
And communication device 45.The quantity of processor 40 can be one or more in the equipment, be with a processor 40 in Fig. 4
Example.The quantity of memory 41 can be one or more in the equipment, in Fig. 4 by taking a memory 41 as an example.The equipment
Processor 40, memory 41, display screen 42, input unit 43, output device 44 and communication device 45 can by bus or
Person's other modes connect, in Fig. 4 for being connected by bus.
Memory 41 is used as a kind of computer readable storage medium, can be used for storing software program, journey can be performed in computer
Sequence and module, the corresponding program instruction of the method for detecting abnormality of the retina as described in any embodiment of that present invention/module (example
Such as, the former retinal images in the abnormal device of above-mentioned retina obtain module 301, picture quality determining module 302, image wind
Lattice transferring module 303 and exception information determining module 304), memory 41 can mainly include storing program area and storage data area,
Wherein, application program needed for storing program area can store operating device, at least one function；Storage data area can store basis
Equipment uses created data etc..In addition, memory 41 may include high-speed random access memory, it can also include non-
Volatile memory, for example, at least a disk memory, flush memory device or other non-volatile solid state memory parts.?
In some examples, memory 41 can further comprise the memory remotely located relative to processor 40, these remote memories
Network connection to equipment can be passed through.The example of above-mentioned network includes but is not limited to internet, intranet, local area network, shifting
Dynamic communication network and combinations thereof.
Display screen 42 is the display screen 42 with touch function, can be capacitance plate, electromagnetic screen or infrared screen.Generally
For, display screen 42 is used to show data according to the instruction of processor 40, is also used to receive the touch behaviour for acting on display screen 42
Make, and corresponding signal is sent to processor 40 or other devices.Optionally, it when display screen 42 is infrared screen, also wraps
Infrared touch frame is included, which is arranged in the surrounding of display screen 42, can be also used for receiving infrared signal, and should
Infrared signal is sent to processor 40 or other equipment.
Communication device 45 communicates to connect for establishing with other equipment, can be wire communication device and/or channel radio
T unit.
Input unit 43 can be used for receiving the number or character information of input, and generate with the user setting of equipment with
And the related key signals input of function control, it can also be the camera for obtaining retinal images and obtain audio data
Pick up facility.Output device 44 may include the audio frequency apparatuses such as loudspeaker.It should be noted that input unit 43 and output dress
Setting 44 concrete composition may be set according to actual conditions.
Software program, instruction and the module that processor 40 is stored in memory 41 by operation, thereby executing equipment
Various function application and data processing, that is, realize the method for detecting abnormality of above-mentioned retina.
Specifically, in embodiment, when processor 40 executes the one or more programs stored in memory 41, specific implementation
The step of method for detecting abnormality of retina provided in an embodiment of the present invention.
Embodiment five
The embodiment of the present invention five additionally provides a kind of computer readable storage medium, is stored thereon with computer program, should
Program can realize the method for detecting abnormality of the retina in any embodiment of that present invention when being executed by processor.This method specifically may be used
To include:
Obtain the former retinal images of a retina；
Determine the picture quality of the former retinal images；
When described image quality meets preset quality requirements, image Style Transfer is carried out to the former retinal images
Pretreatment, obtains targeted retinal image；
The exception information of the retina is determined according to the targeted retinal image.
Certainly, a kind of storage medium comprising computer executable instructions, computer provided by the embodiment of the present invention
The method operation that executable instruction is not limited to the described above, can also be performed the present invention and is mentioned applied to any embodiment in equipment
Relevant operation in the method for detecting abnormality of the retina of confession.
It should be noted that for device, equipment, storage medium embodiment, since it is basic with embodiment of the method
Similar, so being described relatively simple, the relevent part can refer to the partial explaination of embodiments of method.
By the description above with respect to embodiment, it is apparent to those skilled in the art that, the present invention
It can be realized by software and required common hardware, naturally it is also possible to which by hardware realization, but in many cases, the former is more
Good embodiment.Based on this understanding, technical solution of the present invention substantially in other words contributes to the prior art
Part can be embodied in the form of software products, which can store in computer readable storage medium
In, floppy disk, read-only memory (Read-Only Memory, ROM), random access memory (Random such as computer
Access Memory, RAM), flash memory (FLASH), hard disk or CD etc., including some instructions are with so that a computer is set
Standby (can be personal computer, server or the network equipment etc.) executes method described in each embodiment of the present invention.
It is worth noting that, in the embodiment of the abnormal detector of above-mentioned retina, included each unit and mould
Block is only divided according to the functional logic, but is not limited to the above division, and is as long as corresponding functions can be realized
It can；In addition, the specific name of each functional unit is also only for convenience of distinguishing each other, the protection model being not intended to restrict the invention
It encloses.
The above description is only a preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For, the invention can have various changes and changes.All any modifications made within the spirit and principles of the present invention are equal
Replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of method for detecting abnormality of retina characterized by comprising
Obtain the former retinal images of a retina；
Determine the picture quality of the former retinal images；
When described image quality meets preset quality requirements, image Style Transfer is carried out to the former retinal images and is located in advance
Reason, obtains targeted retinal image；
The exception information of the retina is determined according to the targeted retinal image.
2. the method as described in claim 1, which is characterized in that the picture quality of the determination former retinal images, packet
It includes:
The former retinal images are handled to determine assessed value, the assessed value is for expressing the former retinal images
Picture quality；
Judge whether the assessed value is preset target value；
If so, determining that described image quality meets preset quality requirements；
If not, it is determined that institute's commentary picture quality does not meet preset quality requirements, returns to the former view for obtaining a retina
The step of nethike embrane image.
3. the method as described in claim 1, which is characterized in that described to meet preset quality requirements in described image quality
When, the pretreatment of image Style Transfer is carried out to the former retinal images, obtains targeted retinal image, comprising:
Determine that image Style Transfer model, described image Style Transfer model are pre-processed for image Style Transfer；
It will be handled in the former retinal images input described image Style Transfer model, to generate targeted retinal figure
Picture.
4. method as claimed in claim 3, which is characterized in that the determining image Style Transfer model, comprising:
Obtain the first training retinal images；
Determine that the first image composer, the second image composer, the first image generator are located in advance for image Style Transfer
Reason, second image composer are used for image restoring style；
Style Transfer will be carried out in the first training retinal images input the first image generator, obtains the second training
Retinal images；
The second training retinal images are inputted and carry out style reduction in second image composer, obtain third training
Retinal images；
Using the first training retinal images, the second training retinal images and third training retinal map
As being adjusted to the first image generator, to obtain image Style Transfer model.
5. method as claimed in claim 4, which is characterized in that described using the first training retinal images, described the
Two training retinal images and third training retinal images are adjusted the first image generator, to obtain
Image Style Transfer model, comprising:
The first retinal structure feature is extracted from the first training retinal images, and trains retina from described second
The second retinal structure feature is extracted in image；
Calculate the first-loss value between the first retinal structure feature and the second retinal structure feature；
Calculate the second penalty values between the first training retinal images and third training retinal images；
The first image generator is adjusted using the first-loss value and second penalty values, to obtain image
Style Transfer model.
6. the method as described in claim 1, which is characterized in that further include:
Determine whether the retina is abnormal based on the targeted retinal image；
When determining the retinal abnormalities, the exception that the retina is determined according to the targeted retinal image is executed
The step of information.
7. as the method according to claim 1 to 6, which is characterized in that described to be determined according to the targeted retinal image
The exception information of the retina, comprising:
The targeted retinal image is inputted respectively in multiple preset predicting abnormality networks and is handled, multiple dtexs are obtained
Vector is levied, each point feature vector is in a kind of abnormal feature for expressing the retina；
Dtex sign Vector Groups are combined into total characteristic vector；
The total characteristic vector is inputted in preset full Connection Neural Network classifier and is handled, multiple value-at-risks are obtained,
Each value-at-risk is in a kind of abnormal risk for expressing the retina.
8. a kind of abnormal detector of retina characterized by comprising
Former retinal images obtain module, for obtaining the former retinal images of a retina；
Picture quality determining module, for determining the picture quality of the former retinal images；
Image Style Transfer module, for when described image quality meets preset quality requirements, to the former retinal map
As carrying out the pretreatment of image Style Transfer, targeted retinal image is obtained；
Exception information determining module, for determining the exception information of the retina according to the targeted retinal image.
9. a kind of equipment, comprising:
One or more processors；
Storage device, for storing one or more programs；
One or more of programs are executed by one or more of processors, so that one or more of processors are realized
Such as the method for detecting abnormality of retina of any of claims 1-7.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
The method for detecting abnormality such as retina of any of claims 1-7 is realized when execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910251114.6A CN109833025A (en) | 2019-03-29 | 2019-03-29 | A kind of method for detecting abnormality of retina, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910251114.6A CN109833025A (en) | 2019-03-29 | 2019-03-29 | A kind of method for detecting abnormality of retina, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109833025A true CN109833025A (en) | 2019-06-04 |
Family
ID=66886617
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910251114.6A Pending CN109833025A (en) | 2019-03-29 | 2019-03-29 | A kind of method for detecting abnormality of retina, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109833025A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275682A (en) * | 2020-01-19 | 2020-06-12 | 上海箱云物流科技有限公司 | Container detection method, device and computer readable storage medium |
| CN111311546A (en) * | 2020-01-19 | 2020-06-19 | 上海箱云物流科技有限公司 | Container detection method, device and computer readable storage medium |
| CN114842342A (en) * | 2022-05-16 | 2022-08-02 | 网思科技股份有限公司 | Method and device for detecting disordered scene based on artificial intelligence and related equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104398234A (en) * | 2014-12-19 | 2015-03-11 | 厦门大学 | Comprehensive ocular surface analyzer based on expert system |
| US9576351B1 (en) * | 2015-11-19 | 2017-02-21 | Adobe Systems Incorporated | Style transfer for headshot portraits |
| CN107209933A (en) * | 2014-08-25 | 2017-09-26 | 新加坡科技研究局 | For assessing retinal images and the method and system of information being obtained from retinal images |
| CN107358055A (en) * | 2017-07-21 | 2017-11-17 | 湖州师范学院 | Intelligent auxiliary diagnosis system |
| CN108324242A (en) * | 2017-12-28 | 2018-07-27 | 东北大学 | A kind of multispectral fundus imaging device and method based on intelligent terminal |
| CN108564127A (en) * | 2018-04-19 | 2018-09-21 | 腾讯科技（深圳）有限公司 | Image conversion method, device, computer equipment and storage medium |
| CN108846793A (en) * | 2018-05-25 | 2018-11-20 | 深圳市商汤科技有限公司 | Image processing method and terminal device based on image style transformation model |
-
2019
- 2019-03-29 CN CN201910251114.6A patent/CN109833025A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107209933A (en) * | 2014-08-25 | 2017-09-26 | 新加坡科技研究局 | For assessing retinal images and the method and system of information being obtained from retinal images |
| CN104398234A (en) * | 2014-12-19 | 2015-03-11 | 厦门大学 | Comprehensive ocular surface analyzer based on expert system |
| US9576351B1 (en) * | 2015-11-19 | 2017-02-21 | Adobe Systems Incorporated | Style transfer for headshot portraits |
| CN107358055A (en) * | 2017-07-21 | 2017-11-17 | 湖州师范学院 | Intelligent auxiliary diagnosis system |
| CN108324242A (en) * | 2017-12-28 | 2018-07-27 | 东北大学 | A kind of multispectral fundus imaging device and method based on intelligent terminal |
| CN108564127A (en) * | 2018-04-19 | 2018-09-21 | 腾讯科技（深圳）有限公司 | Image conversion method, device, computer equipment and storage medium |
| CN108846793A (en) * | 2018-05-25 | 2018-11-20 | 深圳市商汤科技有限公司 | Image processing method and terminal device based on image style transformation model |
Non-Patent Citations (2)
| Title |
|---|
| TALHA IQBAL，HAZRAT ALI: "Generative Adversarial Network for Medical Images (MI-GAN)", 《JOURNAL OF MEDICAL SYSTEMS》 * |
| 丁蓬莉，李清勇 ，张振，李峰: "糖尿病性视网膜图像的深度神经网络分类方法", 《计算机应用》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275682A (en) * | 2020-01-19 | 2020-06-12 | 上海箱云物流科技有限公司 | Container detection method, device and computer readable storage medium |
| CN111311546A (en) * | 2020-01-19 | 2020-06-19 | 上海箱云物流科技有限公司 | Container detection method, device and computer readable storage medium |
| CN114842342A (en) * | 2022-05-16 | 2022-08-02 | 网思科技股份有限公司 | Method and device for detecting disordered scene based on artificial intelligence and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7075085B2 (en) | Systems and methods for whole body measurement extraction | |
| US11526983B2 (en) | Image feature recognition method and apparatus, storage medium, and electronic apparatus | |
| KR102014377B1 (en) | Method and apparatus for surgical action recognition based on learning | |
| CN111291841B (en) | Image recognition model training method and device, computer equipment and storage medium | |
| JP2021536057A (en) | Lesion detection and positioning methods, devices, devices, and storage media for medical images | |
| CN108171152A (en) | Deep learning human eye sight estimation method, equipment, system and readable storage medium storing program for executing | |
| CN109833025A (en) | A kind of method for detecting abnormality of retina, device, equipment and storage medium | |
| US20070258656A1 (en) | Method, system and computer program product for automatic and semi-automatic modification of digital images of faces | |
| CN108875452A (en) | Face identification method, device, system and computer-readable medium | |
| WO2007128117A1 (en) | Method. system and computer program product for automatic and semi-automatic modification of digital images of faces | |
| CN111931585A (en) | Classroom concentration degree detection method and device | |
| WO2021047185A1 (en) | Monitoring method and apparatus based on facial recognition, and storage medium and computer device | |
| US20220254134A1 (en) | Region recognition method, apparatus and device, and readable storage medium | |
| CN108805058A (en) | Target object changes gesture recognition method, device and computer equipment | |
| CN110298279A (en) | A kind of limb rehabilitation training householder method and system, medium, equipment | |
| US20220366244A1 (en) | Modeling Human Behavior in Work Environment Using Neural Networks | |
| CN106204597A (en) | A kind of based on from the VS dividing method walking the Weakly supervised study of formula | |
| Zhang et al. | A human-in-the-loop deep learning paradigm for synergic visual evaluation in children | |
| CN112257645B (en) | Method and device for positioning key points of face, storage medium and electronic device | |
| CN113421243A (en) | Method and device for detecting type of fundus image based on knowledge distillation network | |
| CN108074240B (en) | Recognition method, recognition apparatus, computer-readable storage medium, and program product | |
| Ray et al. | Design and implementation of affective e-learning strategy based on facial emotion recognition | |
| US10755088B2 (en) | Augmented reality predictions using machine learning | |
| CN113902743A (en) | Method and device for identifying diabetic retinopathy based on cloud computing | |
| CN114372985A (en) | Diabetic retinopathy focus segmentation method and system adapting to multi-center image |
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 |
Application publication date: 20190604 |
|
| RJ01 | Rejection of invention patent application after publication |