CN109464120A - A kind of screening for diabetic retinopathy method, apparatus and storage medium - Google Patents
A kind of screening for diabetic retinopathy method, apparatus and storage medium Download PDFInfo
- Publication number
- CN109464120A CN109464120A CN201811288905.8A CN201811288905A CN109464120A CN 109464120 A CN109464120 A CN 109464120A CN 201811288905 A CN201811288905 A CN 201811288905A CN 109464120 A CN109464120 A CN 109464120A
- Authority
- CN
- China
- Prior art keywords
- eye
- picture
- model
- diabetic retinopathy
- mydriasis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/14—Arrangements specially adapted for eye photography
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Medical Informatics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Ophthalmology & Optometry (AREA)
- Heart & Thoracic Surgery (AREA)
- Computing Systems (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The invention discloses a kind of screening for diabetic retinopathy method, apparatus and storage mediums.Wherein screening for diabetic retinopathy method specifically includes: exempting from mydriasis eye-ground photography pictures, picture pretreatment, modelling, model training and verifying and result judgement serial of methods by constructing double vision open country, one depth convolutional neural networks model of building in advance, when wanting the diabetic retinopathy result of screening patient, first obtain the eye-ground photography picture for exempting from mydriasis to the double vision open country of screening person, then the picture to conform to quality requirements is input in the convolutional neural networks model constructed in advance, so that it may simply, easily obtain diabetic retinopathy result.Compared with prior art, the screening method is simple, convenient, compliance advantage that is good, while having accuracy rate, sensibility and specificity high.
Description
Technical field
The present invention relates to a kind of screening for diabetic retinopathy technologies, and in particular to a kind of diabetic retinopathy sieve
Checking method, device and storage medium.
Background technique
Diabetic retinopathy (Diabetic Retinopathy, DR) is the most common microvascular complication of diabetes
One of, it is the third-largest diseases causing blindness in China.The early treatment of DR can be to prevent serious vision loss in 90% degree.But
Since the complexity of the insufficient and conventional funduscopy of medical resource causes patient compliance poor, up to 50% diabetes
Patient carries out regularly examination of eyes there is no annual.There are many diagnostic method of DR, including funduscopy, fundus fluorescence blood vessel
Radiography, eye-ground photography, seven visual field Color stereoscopic photography of standard (seven standard field 35-mm stereoscopic
Color retinal imaging, 7SF) etc..Wherein 7SF sensibility is strong, and it is internationally recognized retinopathy diagnosis and treatment that diagnosis rate is high
In " goldstandard ".But due to the complexity of 7SF diagnostic method, and patient must inject mydriatic, thus its clinical application is limited；
Eye-ground photography includes haplopia open country and the photograph of more visuals field, convenient for patient's receiving, and can have low cost without using chemical mydriatic,
Many advantages, such as convenient, is most widely used DR diagnostic means, also produces the image data of magnanimity.DR screening at present is one
A time-consuming and laborious artificial process, patient check after after general two days doctor can just submit inspection result, it is such postpone it is past
Toward the cost that will increase doctor patient communication.
Eye-ground photography is the DR screening means that clinical guidelines are recommended, and exempting from mydriasis eye-ground photography using double vision open country can achieve DR
Screening standard (DR screening method index as defined in British Diabetes Association: sensibility: 80%, specificity: 95%).It is opposite to take
When effort mydriasis eye-ground photography technology, exempting from mydriasis eye-ground photography has the characteristics that simple, convenient, compliance is good, is more suitable for making
For the technology of screening.Deep learning model achieves very on the diagnostic imaging of fundus oculi disease diagnostic imaging and Other diseases
High accuracy rate, sensibility and specificity are expected to be applied to efficiency that is clinical and improving clinical diagnosis.
Summary of the invention
It is a time-consuming expense the present invention solves the technical problem of screening for diabetic retinopathy in current clinic
The artificial process of power, image data magnanimity, meanwhile, patient cannot obtain inspection result in time.
In order to solve the above technical problems, the present invention proposes a kind of screening for diabetic retinopathy method, comprising: obtain double
Exempt from the eye-ground photography picture of mydriasis in the visual field；According to the eye-ground photography picture, diabetic retinopathy result is obtained.
A kind of specific embodiment according to the present invention, the present invention also provides a kind of dresses of screening for diabetic retinopathy
It sets, comprising: the eye-ground photography acquisition device of mydriasis is exempted from double vision open country, and the eyeground for exempting from mydriasis for acquiring the double vision open country to screening person is shone
Phase picture；Processor, for executing method as described above.
A kind of specific embodiment according to the present invention, the present invention also provides a kind of computer readable storage mediums, including
Program, described program can be executed by processor to realize method as described above.
According to a kind of screening for diabetic retinopathy method and apparatus of above-described embodiment, exempt to dissipate by obtaining double vision open country
The eye-ground photography picture of pupil, the screening using convolutional neural networks model, for diabetic retinopathy.Screening method letter
Single, convenient, compliance advantage that is good, while having accuracy rate, sensibility and specificity high.
Detailed description of the invention
Fig. 1 is a kind of screening for diabetic retinopathy method flow diagram；
Fig. 2 is a kind of flow chart for constructing screening for diabetic retinopathy convolutional neural networks model method；
Fig. 3 is the flow chart of verifying and result judgement method in a kind of convolutional neural networks model；
Fig. 4 is a kind of screening for diabetic retinopathy device.
Specific embodiment
Below by specific embodiment combination attached drawing, invention is further described in detail.Wherein different embodiments
Middle similar component uses associated similar element numbers.In the following embodiments, many datail descriptions be in order to
The application is better understood.However, those skilled in the art can recognize without lifting an eyebrow, part of feature
It is dispensed, or can be substituted by other elements, material, method in varied situations.In some cases, this Shen
Please it is relevant it is some operation there is no in the description show or describe, this is the core in order to avoid the application by mistake
More descriptions are flooded, and to those skilled in the art, these relevant operations, which are described in detail, not to be necessary, they
Relevant operation can be completely understood according to the general technology knowledge of description and this field in specification.
It is formed respectively in addition, feature described in this description, operation or feature can combine in any suitable way
Kind embodiment.Meanwhile each step in method description or movement can also can be aobvious and easy according to those skilled in the art institute
The mode carry out sequence exchange or adjustment seen.Therefore, the various sequences in the description and the appended drawings are intended merely to clearly describe a certain
A embodiment is not meant to be necessary sequence, and wherein some sequentially must comply with unless otherwise indicated.
It is herein component institute serialization number itself, such as " first ", " second " etc., is only used for distinguishing described object,
Without any sequence or art-recognized meanings.And " connection ", " connection " described in the application, unless otherwise instructed, include directly and
It is indirectly connected with (connection).
Diabetic retinopathy (Diabetic Retinopathy, DR) is the most common microvascular complication of diabetes
One of, according to statistics, DR illness rate is 24.7%-37.5%, and diabetes diagnosis patient in China's is about that 1.1 hundred million, DR patient is super at present
Cross 27,000,000.WESDR (epidemiological study of the Wisconsin diabetic retinopathy) discovery that the world attractes attention, disease
In patient of the journey greater than 15 years, 97% Type I diabetes patient, 80% use the patients with NIDDM and 55% of insulin not
Using the patients with NIDDM of insulin with retinopathy.American Academy of Ophthalmology's guide is also shown, the early treatment of DR
It can be to prevent serious vision loss in 90% degree.
In embodiments of the present invention, a kind of screening for diabetic retinopathy method is proposed, is exempted from by obtaining double vision open country
The eye-ground photography picture of mydriasis, the screening using convolutional neural networks model, for diabetic retinopathy.
Embodiment one:
Referring to FIG. 1, a kind of screening for diabetic retinopathy method, comprising:
Step A001: the eye-ground photography picture that mydriasis is exempted from double vision open country is obtained；
Step A002: according to the eye-ground photography picture, diabetic retinopathy result is obtained.
Wherein, in the present embodiment, step A001: obtaining the eye-ground photography picture that mydriasis is exempted from double vision open country, and acquired is double
The eye-ground photography picture that mydriasis is exempted from the visual field specifically includes: left eye eye fundus image and right eye eyeground figure centered on optic disk and macula lutea
Picture.In the prior art, under normal circumstances, the eyeground photograph that mydriasis is exempted from double vision open country is first carried out after darkroom adapts to 5min after screening person
Mutually take pictures, center of taking pictures is left eye after optic disk and macula lutea elder generation right eye, to the camera site of every eye eyeground picture, clarity, can
Reading scope and total quality are assessed, it is desirable that position is correct in picture, focusing is clear, exposure is moderate.If mixed without interstitial
Turbid image, it is desirable that can clearly show the structure of retina and optic disk, it may be assumed that retinal blood tube wall, retinal nerve fiber
Layer, optic disk disk sieve-plate structure along surface capilary and optic cup are high-visible.The matter of all eyeground pictures is checked after shooting immediately
Amount, picture undesirable for quality are deleted and are re-shoot.After required picture is completed in shooting, export is correct
Format is saved with full resolution.The picture file type saved includes: the tiff format or lossless compression of lossless compression
JPG format.
In the present embodiment, as shown in Fig. 2, step A002: according to the eye-ground photography picture, obtaining diabetic retina
Lesion is as a result, specifically include: one convolution neural network model of building in advance；The eye-ground photography picture is input to the convolution
Neural network model, to obtain diabetic retinopathy result.
Construction method for the convolutional neural networks model is comprising steps of mydriasis eye-ground photography figure is exempted from building double vision open country
Piece collection, picture pretreatment, modelling, model training and verifying and result judgement.
Wherein, step A211: exempt from mydriasis eye-ground photography pictures in building double vision open country, comprising: collect different type diabetes
Exempt from mydriasis eyeground picture and be marked in patient's double vision open country in retina different lesions stage, wherein the fundus photograph after label
Pictures are defined as training dataset.It includes: to assess first picture quality that patient's eyes negative film, which is marked, judgement
Whether can be used for by stages；Secondly, according to international clinic DR severity staging scale (International Clinical
Diabetic Retinopathy scale) to can picture by stages marked by stages, wherein respectively correspond for 1-5 grades: just
Normal eyeground, slight, moderate, severe and proliferation period are labeled as having hard exudate apart from 1, macula lutea center disc diameter
Referable DME (Diabetic macular edema, abbreviation DME, diabetic macular edema).
Step A212: picture pretreatment, comprising: picture is stretched, is rotated, changes brightness and the increased behaviour of contrast
Make, for increasing the complexity of picture, meanwhile, the size of unified all pictures.Exempt from mydriasis eyeground figure obtaining double vision open country at present
During piece, the picture size format of separate sources is different and generally large, by picture pre-treatment step, can reduce ruler
It is very little to be used to reduce memory, and it is conveniently used for batch processing；Meanwhile can to increase model subsequent multiple to clinic for the complexity for increasing picture
The identifying processing ability of miscellaneous picture improves accuracy rate.In the present embodiment, it is big to be unified for 512*512*3 for the size of all pictures
It is small.
Step A213: modelling specifically includes: building convolutional neural networks (CNN) model, the convolutional neural networks
Model includes 12 layers of convolutional layer, 4 layers of pond layer, for extracting 1024 features of image；Wherein, the convolution kernel size used is equal
For 3*3, Filling power (padding) is " SAME "；Chi Huahe is 3*3, step-length 2, for halving the picture size of Chi Huahou；
The core of the last layer convolutional layer is having a size of 8*8, for exporting the feature of all extractions；Full articulamentum totally 1024 nodes,
The probability of dropout is 0.5, and by softmax layer, output category is general for diabetic retinopathy difference result by stages
Rate.
Step A214: model training specifically includes: due to being classification task, using intersection entropy function as the training stage
Loss function, use the ginseng of stochastic gradient descent method (Stochastic gradient descent, SGD) Lai Gengxin network
Number, learning rate are set to 0.01；Every time when training, takes 50 samples to be trained at random in training data concentration, train 2000 altogether
It is secondary.Model training step respectively carries out two visuals field, and particular content includes: first with the training data in first visual field
It obtains the parameter of convolutional neural networks model, and trained model is known as model 1；Then it on the basis of model 1, utilizes
Training data in second visual field continues the update training for carrying out parameter to model 1, and final mask is known as model 2.
Step A215: verifying and result judgement, specifically include: verification step, using with the nonoverlapping eye of training dataset
Negative film collection (referred to as verifying collection) is verified trained model, parameter parameter, it may be assumed that the output knot of comparison model
The consistency of the label result of fruit and picture itself, obtains model index parameter.In the present embodiment, model index parameter packet
It includes: susceptibility, specificity and AUC.Judgment step respectively carries out single eye, i.e., provides and sentence respectively to two eyes of same people
Break as a result, carrying out data processing by the judging result to described two visuals field obtains final judging result, for reducing it
In a visual field image in noise or interference information bring error.In the present embodiment, it to same one eye, will test first
Input of first visual field picture concentrated as model 1 is demonstrate,proved, output score s is calculated1；Then second verifying concentrated
Input of the visual field picture as model 2 calculates output score s2；Data processing is carried out to the judging result in two visuals field, it may be assumed thatWherein, final output score s is the output score s in two visuals field1And s2Arithmetic square root after multiplication, when
When s > γ, it is determined as diabetic retinopathy, is otherwise determined as normal, wherein γ is threshold value, and γ is needed according to model
Output obtains, and specific method is constantly to adjust γ value according to verification result during model training, take verification result preferably when
γ value be final threshold value.
The output result that the method for step A215 as shown in Figure 3 can integrate two visuals field well is (especially different
Spend biggish output result), for example, if s1=0.2, s2The larger situation of the output scoring distinctions in=0.8 this two visuals field
Under, being utilized respectively judgement result that the output score in two visuals field obtains may be completely on the contrary, and output in the present embodiment
It is scored at s=0.4, has more comprehensively reacted the information in two visuals field, can be reduced very well in the image in one of visual field
Noise or interference information bring error.When any one eye of same people is judged as diabetic retinopathy, all think
It is the patient of the disease.The simple eye decision principle not only reduces the risk failed to judge, and meets and be likely to occur two in reality
The nonsynchronous actual conditions of eyes progression of the disease.
In the present embodiment, it when to obtain when the diabetic retinopathy result of screening person, first obtains corresponding double
The eye-ground photography picture of mydriasis is exempted from the visual field, then the picture to conform to quality requirements is input to the convolutional neural networks model of building
In, so that it may simply, easily obtain diabetic retinopathy result.
Embodiment two:
Referring to FIG. 4, a kind of screening for diabetic retinopathy device, comprising: exempt from mydriasis eye-ground photography acquisition in double vision open country
Mydriasis eye-ground photography picture is exempted from device B00, the double vision open country for acquiring to screening person；Processor B10, for executing a kind of glycosuria
The method of sick retinopathy screening.Wherein, double vision open country exempt from mydriasis eye-ground photography acquisition device B00 can be with processor B10 collection
At together, screening for diabetic retinopathy device can also be formed respectively as individual components.
In the present embodiment, a kind of screening for diabetic retinopathy method performed by processor B10 include: obtain it is double
Exempt from the eye-ground photography picture of mydriasis in the visual field；According to the eye-ground photography picture, diabetic retinopathy result is obtained.Wherein,
The eye-ground photography picture that mydriasis is exempted from the double vision open country includes left eye eye fundus image and right eye eyeground centered on optic disk and macula lutea
Image.
In the present embodiment, described according to the eye-ground photography picture, obtain diabetic retinopathy result, comprising:
One convolution neural network model of building in advance；The eye-ground photography picture is input to the convolutional neural networks model, to obtain
Take diabetic retinopathy result.Wherein, the construction method of the convolutional neural networks model is comprising steps of building double vision is wild
Exempt from mydriasis eye-ground photography pictures, picture pretreatment, modelling, model training and verifying and result judgement.
In the present embodiment, it includes: to collect different type glycosuria that mydriasis eye-ground photography pictures are exempted from the building double vision open country
Exempt from mydriasis eyeground picture and be marked in patient's double vision open country in sick retina different lesions stage, wherein the eyeground after label is shone
Piece pictures are defined as training dataset；The picture pre-treatment step includes: to be stretched, rotated to picture, changing brightness
And the increased operation of contrast, for increasing the complexity of picture, meanwhile, the size of unified all pictures.The modelling
Step includes: building convolutional neural networks model, and the convolutional neural networks model includes 12 layers of convolutional layer, 4 layers of pond layer, is used
In 1024 features for extracting image；Wherein, the convolution kernel size used is 3*3, and Filling power is " SAME "；Chi Huahe is 3*
3, step-length 2, for halving the picture size of Chi Huahou；The core of the last layer convolutional layer is all for exporting having a size of 8*8
The feature of extraction；Totally 1024 nodes, the probability of dropout are 0.5 to full articulamentum, and by softmax layers, output category is sugar
Urinate the probability of sick retinopathy difference result by stages.The model training step includes: using intersection entropy function as training
The loss function in stage, the parameter of network is updated using stochastic gradient descent method, and learning rate is set to 0.01；Every time when training,
It takes 50 samples to be trained at random in training data concentration, trains 2000 times altogether；The model training step is respectively to two
The visual field carries out: obtaining the parameter of convolutional neural networks model first with the training data in first visual field, obtains model 1；
Then on the basis of model 1, using the training data in second visual field, continue the update training that parameter is carried out to model 1,
Obtain model 2.
In the present embodiment, the verifying and result judgement step include: verification step, are not weighed using with training dataset
Folded double vision open country is exempted from mydriasis eyeground pictures (referred to as verifying collection) and is verified to trained model, parameter parameter,
That is: the consistency of the label result of the output result and picture itself of comparison model, obtains model index parameter.In the present embodiment
In, model index parameter includes: susceptibility, specificity and AUC.Judgment step respectively carries out single eye, i.e., to same people's
Two eyes provide judging result respectively, carry out data processing by the judging result to described two visuals field and obtain final judgement
As a result, for reducing noise or interference information bring error in the image in one of visual field.In the present embodiment, to same
One eye, first visual field picture for first concentrating verifying calculate output score s as the input of model 11；Then will
Input of second visual field picture concentrated as model 2 is verified, output score s is calculated2；To the judging result in two visuals field
Carry out data processing, it may be assumed thatWherein, final output score s is the output score s in two visuals field1And s2It is multiplied it
Arithmetic square root afterwards is determined as diabetic retinopathy as s > γ, is otherwise determined as normal, wherein γ is threshold value,
γ needs are obtained according to the output of model, and specific method is constantly to adjust γ value according to verification result during model training,
Take verification result preferably when γ value be final threshold value.
In the present embodiment, when the diabetes using screening for diabetic retinopathy device acquisition to screening person regard
When retinopathy result, first passes through double vision open country and exempt from mydriasis eye-ground photography acquisition device B00 acquisition and exempt to dissipate to the double vision open country of screening person
The picture to conform to quality requirements, is then input in processor B10 and executes a kind of diabetic retina by pupil eye-ground photography picture
The method of lesion screening simply, can be obtained easily by the convolutional neural networks model constructed in advance in processor B10
Diabetic retinopathy result.
It will be understood by those skilled in the art that all or part of function of various methods can pass through in above embodiment
The mode of hardware is realized, can also be realized by way of computer program.When function all or part of in above embodiment
When being realized by way of computer program, which be can be stored in a computer readable storage medium, and storage medium can
To include: read-only memory, random access memory, disk, CD, hard disk etc., it is above-mentioned to realize which is executed by computer
Function.For example, program is stored in the memory of equipment, when executing program in memory by processor, can be realized
State all or part of function.In addition, when function all or part of in above embodiment is realized by way of computer program
When, which also can store in storage mediums such as server, another computer, disk, CD, flash disk or mobile hard disks
In, through downloading or copying and saving into the memory of local device, or version updating is carried out to the system of local device, when logical
When crossing the program in processor execution memory, all or part of function in above embodiment can be realized.
Using a kind of screening for diabetic retinopathy method and apparatus proposed by the present invention, builds up and exempted from based on double vision open country
The convolutional neural networks model of mydriasis eye-ground photography, and the result that the artificial intelligence model is used for diabetic retinopathy is sieved
Cha Zhong so as to become simple, convenient, compliance good for screening method, while having high excellent of accuracy rate, sensibility and specificity
Gesture.
Use above specific case is illustrated the present invention, is merely used to help understand the present invention, not to limit
The system present invention.For those skilled in the art, according to the thought of the present invention, can also make several simple
It deduces, deform or replaces.
Claims (10)
1. a kind of screening for diabetic retinopathy method characterized by comprising
Obtain the eye-ground photography picture that mydriasis is exempted from double vision open country；
According to the eye-ground photography picture, diabetic retinopathy result is obtained.
2. the method as described in claim 1, which is characterized in that the eye-ground photography picture that mydriasis is exempted from the double vision open country includes to regard
Left eye eye fundus image and right eye eye fundus image centered on disk and macula lutea.
3. the method as described in claim 1, which is characterized in that it is described according to the eye-ground photography picture, obtain diabetes view
Retinopathy result, comprising:
One convolution neural network model of building in advance；
The eye-ground photography picture is input to the convolutional neural networks model, to obtain diabetic retinopathy result.
4. method as claimed in claim 3, which is characterized in that the construction method of the convolutional neural networks model includes step
It is rapid: to exempt from mydriasis eye-ground photography pictures, picture pretreatment, modelling, model training and verifying and sentence with result in building double vision open country
It is fixed.
5. method as claimed in claim 4, which is characterized in that exempt from mydriasis eye-ground photography pictures packet in the building double vision open country
It includes: exempting from mydriasis eye-ground photography picture and go forward side by side rower in the patient's double vision open country for collecting the different type diabetic retina different lesions stage
Note, wherein the fundus photograph pictures after label are defined as training dataset；The picture pre-treatment step includes: to picture
It stretched, rotated, changing brightness and the increased operation of contrast, for increasing the complexity of picture, meanwhile, unified all figures
The size of piece.
6. method as claimed in claim 5, which is characterized in that the modelling step includes: building convolutional neural networks
Model, the convolutional neural networks model includes 12 layers of convolutional layer, 4 layers of pond layer, for extracting 1024 features of image；Its
In, the convolution kernel size used is 3*3, and Filling power is " SAME "；Chi Huahe is 3*3, step-length 2, for by Chi Huahou's
Picture size halves；The core of the last layer convolutional layer is having a size of 8*8, for exporting the feature of all extractions；Full articulamentum is total
1024 nodes, the probability of dropout are 0.5, and by softmax layers, output category is diabetic retinopathy difference point
The probability of phase result.
7. method as claimed in claim 5, which is characterized in that the model training step includes: to be made using intersection entropy function
For the loss function of training stage, the parameter of network is updated using stochastic gradient descent method, learning rate is set to 0.01；Instruction every time
When practicing, takes 50 samples to be trained at random in training data concentration, train 2000 times altogether；The model training step is right respectively
Two visuals field carry out: obtaining the parameter of convolutional neural networks model first with the training data in first visual field, obtain mould
Type 1；Then on the basis of model 1, using the training data in second visual field, continue the update that parameter is carried out to model 1
Training, obtains model 2.
8. method as claimed in claim 5, which is characterized in that described verify with result judgement step includes: verification step, is adopted
Trained model is verified with mydriasis eyeground pictures are exempted from the nonoverlapping double vision open country of training dataset, parameter ginseng
Number；Judgment step respectively carries out single eye, i.e., judging result is provided respectively to two eyes of same people, by described two
The judging result in a visual field carries out data processing and obtains final judging result, in the image for reducing one of visual field
Noise or interference information bring error；The model index parameter that the verification step calculates include: susceptibility, specificity and
AUC；In the judgment step, to same one eye, first by with first in the nonoverlapping eyeground picture library of training dataset
Input of the visual field picture as model 1 calculates output score s1；It then will be with the nonoverlapping eyeground picture library of training dataset
In input of second visual field picture as model 2, calculate output score s2；The judging result in two visuals field is counted
According to processing, it may be assumed thatWherein, final output score s is the output score s in two visuals field1And s2Calculation after multiplication
Art square root is determined as diabetic retinopathy as s > γ, is otherwise determined as normal, wherein γ is threshold value.
9. a kind of device of screening for diabetic retinopathy characterized by comprising
The eye-ground photography acquisition device of mydriasis is exempted from double vision open country, exempts from the eye-ground photography figure of mydriasis for acquiring the double vision open country to screening person
Piece；
Processor, for executing such as method of any of claims 1-8.
10. a kind of computer readable storage medium, which is characterized in that including program, described program can be executed by processor with
Realize such as method of any of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811288905.8A CN109464120A (en) | 2018-10-31 | 2018-10-31 | A kind of screening for diabetic retinopathy method, apparatus and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811288905.8A CN109464120A (en) | 2018-10-31 | 2018-10-31 | A kind of screening for diabetic retinopathy method, apparatus and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109464120A true CN109464120A (en) | 2019-03-15 |
Family
ID=65666728
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811288905.8A Pending CN109464120A (en) | 2018-10-31 | 2018-10-31 | A kind of screening for diabetic retinopathy method, apparatus and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109464120A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110164552A (en) * | 2019-05-27 | 2019-08-23 | 苏州嘉展科技有限公司 | A kind of system and method for extensive screening diabetic and tracking treatment effect |
| CN110956628A (en) * | 2019-12-13 | 2020-04-03 | 广州达安临床检验中心有限公司 | Picture grade classification method and device, computer equipment and storage medium |
| CN112053780A (en) * | 2020-08-24 | 2020-12-08 | 宁波市眼科医院 | Artificial intelligence cataract screening referral system based on eye ground picture |
| CN112581448A (en) * | 2020-12-17 | 2021-03-30 | 中南大学 | Retina image hard exudate identification method and imaging method |
| CN112869706A (en) * | 2021-02-24 | 2021-06-01 | 复旦大学 | Double-view eyeground image fusion method based on deep learning |
| CN112890764A (en) * | 2021-01-18 | 2021-06-04 | 哈尔滨工业大学 | Unmanned low-cost portable fundus disease detection method |
| CN115620899A (en) * | 2022-12-01 | 2023-01-17 | 首都医科大学附属北京朝阳医院 | Multi-visual field-based 3D convolutional neural network integrated model lung nodule diagnosis system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934798A (en) * | 2017-02-20 | 2017-07-07 | 苏州体素信息科技有限公司 | Diabetic retinopathy classification stage division based on deep learning |
| CN107330876A (en) * | 2017-06-12 | 2017-11-07 | 济南浪潮高新科技投资发展有限公司 | A kind of image automatic diagnosis method based on convolutional neural networks |
| CN107423571A (en) * | 2017-05-04 | 2017-12-01 | 深圳硅基仿生科技有限公司 | Diabetic retinopathy identifying system based on eye fundus image |
| CN108095683A (en) * | 2016-11-11 | 2018-06-01 | 北京羽医甘蓝信息技术有限公司 | The method and apparatus of processing eye fundus image based on deep learning |
| CN108154505A (en) * | 2017-12-26 | 2018-06-12 | 四川大学 | Diabetic retinopathy detection method and device based on deep neural network |
| CN108231194A (en) * | 2018-04-04 | 2018-06-29 | 苏州医云健康管理有限公司 | A kind of disease diagnosing system |
| CN108665457A (en) * | 2018-05-16 | 2018-10-16 | 腾讯科技（深圳）有限公司 | Image-recognizing method, device, storage medium and computer equipment |
-
2018
- 2018-10-31 CN CN201811288905.8A patent/CN109464120A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108095683A (en) * | 2016-11-11 | 2018-06-01 | 北京羽医甘蓝信息技术有限公司 | The method and apparatus of processing eye fundus image based on deep learning |
| CN106934798A (en) * | 2017-02-20 | 2017-07-07 | 苏州体素信息科技有限公司 | Diabetic retinopathy classification stage division based on deep learning |
| CN107423571A (en) * | 2017-05-04 | 2017-12-01 | 深圳硅基仿生科技有限公司 | Diabetic retinopathy identifying system based on eye fundus image |
| CN107330876A (en) * | 2017-06-12 | 2017-11-07 | 济南浪潮高新科技投资发展有限公司 | A kind of image automatic diagnosis method based on convolutional neural networks |
| CN108154505A (en) * | 2017-12-26 | 2018-06-12 | 四川大学 | Diabetic retinopathy detection method and device based on deep neural network |
| CN108231194A (en) * | 2018-04-04 | 2018-06-29 | 苏州医云健康管理有限公司 | A kind of disease diagnosing system |
| CN108665457A (en) * | 2018-05-16 | 2018-10-16 | 腾讯科技（深圳）有限公司 | Image-recognizing method, device, storage medium and computer equipment |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110164552A (en) * | 2019-05-27 | 2019-08-23 | 苏州嘉展科技有限公司 | A kind of system and method for extensive screening diabetic and tracking treatment effect |
| CN110956628A (en) * | 2019-12-13 | 2020-04-03 | 广州达安临床检验中心有限公司 | Picture grade classification method and device, computer equipment and storage medium |
| CN112053780A (en) * | 2020-08-24 | 2020-12-08 | 宁波市眼科医院 | Artificial intelligence cataract screening referral system based on eye ground picture |
| CN112581448A (en) * | 2020-12-17 | 2021-03-30 | 中南大学 | Retina image hard exudate identification method and imaging method |
| CN112890764A (en) * | 2021-01-18 | 2021-06-04 | 哈尔滨工业大学 | Unmanned low-cost portable fundus disease detection method |
| CN112890764B (en) * | 2021-01-18 | 2022-12-13 | 哈尔滨工业大学 | Unmanned low-cost portable eye ground disease detection system |
| CN112869706A (en) * | 2021-02-24 | 2021-06-01 | 复旦大学 | Double-view eyeground image fusion method based on deep learning |
| CN115620899A (en) * | 2022-12-01 | 2023-01-17 | 首都医科大学附属北京朝阳医院 | Multi-visual field-based 3D convolutional neural network integrated model lung nodule diagnosis system |
| CN115620899B (en) * | 2022-12-01 | 2023-04-07 | 首都医科大学附属北京朝阳医院 | Multi-view field based 3D convolutional neural network integrated model pulmonary nodule diagnostic system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109464120A (en) | A kind of screening for diabetic retinopathy method, apparatus and storage medium | |
| US11210789B2 (en) | Diabetic retinopathy recognition system based on fundus image | |
| CN109635862B (en) | Sorting method for retinopathy of prematurity plus lesion | |
| CN113646805A (en) | Image-based detection of ophthalmic and systemic diseases | |
| US11197608B2 (en) | Eye disease diagnosis method and system using artificial intelligence | |
| Bader Alazzam et al. | [Retracted] Identification of Diabetic Retinopathy through Machine Learning | |
| US20210391056A1 (en) | Health big data service method and system based on remote fundus screening | |
| CN109998477A (en) | A kind of high myopia cataract operation prognosis intelligence anticipation system | |
| CN112101424B (en) | Method, device and equipment for generating retinopathy identification model | |
| CN109036556A (en) | A method of keratoconus case is diagnosed based on machine learning | |
| Wang et al. | Progress of artificial intelligence in diabetic retinopathy screening | |
| Ram et al. | The relationship between Fully Connected Layers and number of classes for the analysis of retinal images | |
| CN113887662A (en) | Image classification method, device, equipment and medium based on residual error network | |
| CN112869697A (en) | Judgment method for simultaneously identifying stage and pathological change characteristics of diabetic retinopathy | |
| Hassan et al. | Exploiting the transferability of deep learning systems across multi-modal retinal scans for extracting retinopathy lesions | |
| CN111700582A (en) | Common ocular surface disease diagnosis system based on intelligent terminal | |
| Jena et al. | Detection of diabetic retinopathy images using a fully convolutional neural network | |
| Triyadi et al. | Deep learning in image classification using vgg-19 and residual networks for cataract detection | |
| Camara et al. | A comprehensive review of methods and equipment for aiding automatic glaucoma tracking | |
| Suedumrong et al. | Application of deep convolutional neural networks vgg-16 and googlenet for level diabetic retinopathy detection | |
| CN113744254B (en) | Fundus image analysis method, fundus image analysis system, storage medium and computer equipment | |
| CN113273959B (en) | Portable diabetic retinopathy diagnosis and treatment instrument | |
| Do et al. | Prediction of Retina Damage in Optical Coherence Tomography Image Using Xception Architecture Model | |
| CN112869706A (en) | Double-view eyeground image fusion method based on deep learning | |
| Sinha et al. | Prospects of artificial intelligence in ophthalmic practice |
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: 20190315 |
|
| RJ01 | Rejection of invention patent application after publication |