CN108665447A - A kind of glaucoma image detecting method based on eye-ground photography deep learning - Google Patents
A kind of glaucoma image detecting method based on eye-ground photography deep learning Download PDFInfo
- Publication number
- CN108665447A CN108665447A CN201810360708.6A CN201810360708A CN108665447A CN 108665447 A CN108665447 A CN 108665447A CN 201810360708 A CN201810360708 A CN 201810360708A CN 108665447 A CN108665447 A CN 108665447A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- glaucoma
- eye
- eye fundus
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
- G06T7/0014—Biomedical image inspection using an image reference approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Abstract
The invention discloses a kind of glaucoma image detecting methods based on eye-ground photography deep learning.The original eye fundus image collected and marked in database is pre-processed, obtains training example eye fundus image, composing training database;It is expanded, the tranining database after being expanded;Foundation includes the convolutional neural networks of multilayer neural network structure, is trained with the tranining database input convolutional neural networks after amplification;For eye fundus image to be measured, eye fundus image to be measured is input in the convolutional neural networks after training, obtains the output valve of last layer of neural network structure, and then differentiate to glaucoma.The present invention can continue to optimize the parameter of data characteristics and convolutional neural networks for judgement, so as to greatly improve accuracy rate, the sensibility and specificity of glaucoma image detection, save medical resources.
Description
Technical field
The present invention relates to a kind of image processing methods, more particularly, to a kind of blueness based on eye-ground photography deep learning
Light eye image detecting method.
Background technology
Glaucoma is one group of threat and damage optic nerve visual performance, mainly increases related eye disease with pathologic intraocular pressure,
Its illness rate rises year by year.According to the data of the World Health Organization, it is listed in the second largest common cause of whole world blindness.By
The progressive stage of disease is usually arrived when going to a doctor in glaucoma human body, therefore glaucoma is otherwise known as noiseless thief.Though
Right glaucoma can not be cured, caused by vision impairment it is irreversible, but early diagnosis, drug therapy appropriate and hand can be passed through
Art prevents the further damage of eyesight.Therefore, the early diagnosis of glaucoma is most important for the incidence for reducing blindness.
In existing glaucoma computer-aided diagnosis technology, generally some projects mark feature based on Manual definition and into
Row.The morphological change of the identification of optic disk optic cup and layer of optic fibers (RNFL) is most important to the auxiliary diagnosis of glaucoma.
Cup disc ratio (CDR) is one of the important parameter for diagnosis of glaucoma, however, the measurement of CDR is frequently subjected to the master of oculist
The property seen influences.Traditional glaucoma computer-aided diagnosis based on morphological feature, there is following defect:First, Manual definition
Feature there is limitation, be unable to fully cause the accuracy in practical application limited using the information in eye fundus image;The
Two, algorithm is static, and accuracy can not be improved with the increase of patient data.
Invention content
In order to solve the problems, such as background technology, the object of the present invention is to provide one kind being based on eye-ground photography depth
The glaucoma image detecting method of habit, can make full use of in eye fundus image image information detection obtain eye fundus image in whether
Reference result with glaucoma obtains glaucoma information for carrying out auxiliary judgment to glaucoma
The technical solution adopted in the present invention is to include the following steps:
Step 1:The original eye fundus image collected and marked in database is pre-processed, obtains training example eyeground
Image, by training example eye fundus image composing training database;
Original eye fundus image be by fundus camera various different people objects are shot to obtain, and be labeled in advance glaucoma and
Two normal classes.
Step 2:Training example eye fundus image in the tranining database is expanded, the training number after being expanded
According to library;In tranining database, glaucoma and bottom of the normal eyes all have the trained example eye fundus image;
Step 3:Include the convolutional neural networks of multilayer neural network structure for glaucoma and bottom of the normal eyes foundation, with expansion
Tranining database input convolutional neural networks after increasing are trained;
Step 4:For eye fundus image to be measured, eye fundus image to be measured is input in the convolutional neural networks after training, is obtained
The output valve of last layer of neural network structure is obtained, and then glaucoma is differentiated.
Pretreatment in the step 1 include image cropping, image enhancement and pixel normalization, to original eye fundus image according to
Secondary progress image cropping, image enhancement and the normalized processing of pixel obtain training example eye fundus image.
In the step 1, glaucoma and normal eyes that the training example eye fundus image in composing training database is carried
The quantitative proportion of bottom classification annotation is 1:1.
The step 2 uses the method for mirror image and rotation and is carried out in the method that fixed dimension is divided in original image sliding window
Amplification.
The step 2 is specially:
2.1) to each Zhang Xunlian examples eye fundus image carry out it is down-sampled obtain the eye fundus image of 256 × 256 resolution ratio, and
With the image upper left corner be starting origin along horizontal axis to the right with establish sliding window under vertical axis, horizontal axis is to the right or vertical axis order side
To sliding window be further to be cut image by sliding window using the random integers between 0-32 as 4 of step-length sliding windows at equal intervals
At 4 × 4=16 block subgraphs so that picture number expands 4 × 4=16 times to step 2 initially;
2.2) vertically and horizontally mirror image, 90 °, 180 ° and 270 ° are used with random ratio to the subgraph that step 2.1) obtains
Five kinds of methods of rotation carry out second and expand respectively, obtain training dataset.
When convolutional neural networks are trained, the volume is repeatedly trained using the correspondence classification based training example eye fundus image
Product neural network framework adjusts the Parameters of Neural Network Structure, after repeatedly being trained according to the learning rate of setting
Convolutional neural networks.
The convolutional neural networks framework includes 14 layers of neuron, respectively is input layer, convolutional layer, maximum pond
Layer, maximum pond layer, convolutional layer, convolutional layer, convolutional layer, maximum pond layer, full articulamentum, abandons layer, full connection at convolutional layer
Layer abandons layer and output layer;Wherein, five layers of convolutional layer C1~C5 are using the letter between ReLU activation primitives simulation biological neuron
Number transmit, preceding two layers of convolutional layer C1 and C2 be respectively adopted size be 11 × 11 and 5 × 5 convolution kernel input picture is rolled up
Product, rear three-layer coil lamination C3~C5 then use 3 × 3 convolution kernel;C1~C5 is convolutional layer and with ReLU (Rectified
Linear Units, correct linear unit) activation primitive simulation biological neuron between signal transmit, C1 and C2 are respectively adopted
The convolution kernel that size is 11 × 11 and 5 × 5 carries out convolution to input picture, and C3~C5 is all made of 3 × 3 convolution kernel, and rolls up
The window that lamination C1, C2, C5 are all made of 3 × 3 after activation primitive activates carries out maximum pond layer operation and passes through local acknowledgement
Regularization (Local Response Normalization, LRN) is connect with back layer grade;Wherein, two layers full articulamentum F1 and
F2 is connected in the form of neuron connects entirely with front and back adjacent layer.
Convolutional neural networks use cross entropy cost function as loss function when training, and use the boarding steps based on momentum
Descent algorithm is spent as study optimization algorithm.
For the glaucoma and bottom of the normal eyes image of doctor's mark, multilayer convolutional neural networks are established, are based on eye fundus image
Training to convolutional neural networks, make the final output value of convolutional neural networks meet doctor mark as a result, so as to profit
Glaucoma is carried out with trained convolutional neural networks to detect auxiliary automatically.The eyeground that the method for the present invention passes through training label
Image is realized feature needed for learning from training example eye fundus image library automatically using deep learning and carries out discriminant classification.
Preferably, when training convolutional neural networks, trained learning rate is less than or equal to the preceding study once trained every time
Rate.
With above-mentioned technical proposal so that the present invention has following advantage compared with prior art:
The present invention can apply to data volume relatively small database and be realized certainly by deep learning after data amplification
It is dynamic to learn required feature from tranining database and carry out discriminant classification.
The present invention can continue to optimize the parameter of data characteristics and convolutional neural networks for judgement, so as to substantially
Improve accuracy rate, the sensibility and specificity of glaucoma image detection, save medical resources.
Description of the drawings
Attached drawing 1 is the flow chart of the method for the present invention.
Specific implementation mode
With reference to embodiment, the present invention will be further described.
The embodiment of the present invention and its implementation process are as follows:
Step 1:The eye fundus image of acquisition came from 2nd Affiliated Hospital Zhejiang University School of Medicine Eye Center, at 2016 8
Month in 10 months of in June, 2017,2095 from 1443 human bodies, the age at 2 years old to 90 years old, the eye-ground photography of shooting
Machine is the TRC-NW8 fundus cameras (TopCon Medical Systems, Tokyo, Japan) of Table top type.By two ophthalmology
Doctor shoots, and eye fundus image resolution ratio is 2144 × 1424 pixels.Shooting eye-ground photography carries out in darkroom, without using scattered
Pupil agent has serious refracting media problem that cannot take eye fundus image, then excludes human body except this research.Glaucoma according to
(National Institute for Health and Clinical are affixed one's name to according to the state-run quality of medical care standard of Britain
Excellence, NICE) guidelines standards mark.
Step 2:Eye fundus image in the database is pre-processed.Eye fundus image is cut into centered on optic disk
The image of the unified size of 356 × 356 resolution ratio, using contrast limitation self-adapting histogram equilibrium (CLAHE) to image into
Row enhancing processing and pixel normalization, obtain training example eye fundus image, and instruction is constituted by the eye fundus image of training example glaucoma
Practice the data of database;
Step 3:The data of the tranining database are expanded.Using vertically and horizontally mirror image, 90o, 180 ° and
The method of 270 ° of rotations, and with fixed dimension a × a in the side that the image sliding window of original size b × b (a < b) size is divided
Method.Tranining database after being expanded by the above method, it is reasonable to be carried out to glaucoma eye fundus image and normal eyes eye fundus image
Multiple expands so that glaucoma and bottom of the normal eyes amount of images 1 after amplification:1.In tranining database, glaucoma and bottom of the normal eyes
The corresponding classification based training example eye fundus image with the trained example eye fundus image;
3.1) to each Zhang Xunlian examples eye fundus image carry out it is down-sampled obtain the eye fundus image of 256 × 256 resolution ratio, and
With the image upper left corner be starting origin along horizontal axis to the right with establish sliding window under vertical axis, horizontal axis is to the right or vertical axis order side
To sliding window be further to be cut image by sliding window using the random integers between 0-32 as 4 of step-length sliding windows at equal intervals
At 4 × 4=16 block subgraphs so that picture number extends to 4 × 4=16 times of step 2;
3.2) to step 3.1) obtain subgraph with specific multiple (according to positive and negative sample proportion in step 2, the multiple
Purpose be so that amplification is finally obtained positive and negative sample size as close possible to 1:1) and with random ratio with vertically and horizontally mirror
Picture, 90 °, 180 ° and 270 ° of rotations, five kinds of methods carry out second and expand respectively, obtain training dataset.
Step 4:Convolutional neural networks are established for glaucoma and bottom of the normal eyes image;The convolutional neural networks include
Multilayer neural network structure.For the convolutional neural networks, repeatedly instructed using the correspondence classification based training example eye fundus image
Practice the neural network framework in the convolutional neural networks, the nerve net is adjusted according to the learning rate 0.001 of setting when training
Network configuration parameters, to obtain the convolutional neural networks after the multiple training of glaucoma;
Step 5:Output based on last layer of neural network framework in the convolutional neural networks after each training
Value differentiates glaucoma.
Convolutional neural networks framework includes 14 layers of neuron, respectively is input layer, convolutional layer, maximum pond layer, volume
Lamination, convolutional layer, convolutional layer, convolutional layer, maximum pond layer, full articulamentum, abandons layer, full articulamentum, discarding at maximum pond layer
Layer and output layer, it is as shown in Figure 1 for the structure of the main frame of the full articulamentum of convolution sum;
Wherein input layer trains the training image example in case library for inputting the eye fundus image pre-processed.Most
Great Chiization layer (max-pooling) is used to be maximized as defeated last output in the sliding window of fixed size
Go out, to carry out dimension-reduction treatment to parameter, reduces follow-up calculation amount and improve calculating speed.Layer (dropout) is abandoned for instructing
Unit during practicing in a part of hidden layer of temporary random drop, prevents that overfitting problem occurs in the training process while dropping
Low parameter dimension mitigates training time load.Output layer is last layer, and by softmax graders, its output valve is 0 or 1:
1 represents detection glaucoma, and glaucoma is not detected in 0 representative.
As shown in Figure 1, five layers of convolutional layer C1~C5 are passed using the signal between ReLU activation primitives simulation biological neuron
It passing, preceding two layers of convolutional layer C1 and C2 are respectively adopted the convolution kernel that size is 11 × 11 and 5 × 5 and carry out convolution to input picture, after
Three-layer coil lamination C3~C5 then uses 3 × 3 convolution kernel;C1~C5 is convolutional layer and with ReLU (Rectified Linear
Units corrects linear unit) signal between activation primitive simulation biological neuron transmits, and C1 and C2 are respectively adopted size and are
11 × 11 and 5 × 5 convolution kernel carries out convolution to input picture, and C3~C5 is all made of 3 × 3 convolution kernel, and convolutional layer C1,
The window that C2, C5 are all made of 3 × 3 after activation primitive activates carries out maximum pond layer operation and by local acknowledgement's regularization
(Local Response Normalization, LRN) is connect with back layer grade, i.e., place is normalized to local input image
The instruction of eye fundus image can be improved to the suppression (lateral inhibition) of adjacent neurons in reason, the biologically active neuron of simulation
Practice speed;Shallow-layer C1 is completed by five layers of convolutional layer respectively to deep layer C2 feature extractions, and be input to subsequent full articulamentum to
" the distributed nature expression " acquired is mapped to sample labeling space and carries out last classification hence into softmax layers.
Wherein, two layers of full articulamentum F1 and F2 is connected in the form of neuron connects entirely with front and back adjacent layer.
The present invention distinguishingly rolls up between first convolutional layer and second convolutional layer, in second convolutional layer and third
It is provided with maximum pond layer between lamination and between last convolutional layer and full connection convolutional layer, improves the instruction of eye fundus image
Practice speed, and discarding layer is added also between two layers of full articulamentum and between the full articulamentum of the latter and output layer, passes through
Dropout operations increase network sparsity, prevent over-fitting.
Convolutional neural networks use cross entropy cost function as loss function when training, and use the boarding steps based on momentum
Descent algorithm is spent as study optimization algorithm.
In specific implementation, the inertia changed by the stochastic gradient descent algorithm analogue data based on momentum is optimizing more
More new direction before retaining to a certain extent when new finely tunes final more new direction simultaneously also by present invention study, to
Increase the stability of study, and with certain ability for breaking away from local optimum, specifically uses following formula:
Δxt=m* Δs xt-1-α*gt
Wherein, Δ xtWith Δ xt-1Respectively t and the displacement at t-1 moment update, and m indicates momentum, and α is learning rate, gtFor t when
The gradient at quarter.
With above-mentioned technical proposal so that the present invention has following advantage compared with prior art:The present invention can apply to
Data volume relatively small database after data amplification, is realized by deep learning automatically from tranining database needed for study
Feature simultaneously carries out discriminant classification, constantly corrects data characteristics and the convolutional neural networks ginseng for judgement in the training process
Number, to improve the sensibility and specificity in clinical application, with the increased number of training example eye fundus image, the standard of classification
True property and reliability also will further improve.
By deep learning, realization learns required feature and classify to sentence from tranining database this method automatically
Not, it is continued to optimize in training process for sentencing another characteristic and parameter.Early period test in, this method used 2095 width by
The eye fundus image of oculist's mark is trained, and compares the artificial judgment of doctor, and the accuracy rate of test reaches 90.8%, sensitivity
Property be 92.1%, specificity be 89.5%.Using present invention could apply to hospital clinical, physical examination screening, tele-medicines above
With the fields such as human body self-test.
Claims (8)
1. a kind of glaucoma image detecting method based on eye-ground photography deep learning, it is characterised in that include the following steps:
Step 1:The original eye fundus image collected and marked in database is pre-processed, obtains training example eyeground figure
Picture, by training example eye fundus image composing training database;
Step 2:Training example eye fundus image in the tranining database is expanded, the training data after being expanded
Library;
Step 3:Include the convolution god of multilayer neural network structure for the foundation of the original eye fundus image of glaucoma and bottom of the normal eyes
Through network, it is trained with the tranining database input convolutional neural networks after amplification;
Step 4:For eye fundus image to be measured, eye fundus image to be measured is input in the convolutional neural networks after training, is obtained most
The output valve of later layer neural network structure, and then glaucoma is differentiated.
2. a kind of glaucoma image detecting method based on eye-ground photography deep learning according to claim 1, feature
It is:Pretreatment in the step 1 includes that image cropping, image enhancement and pixel normalize, successively to original eye fundus image
Image cropping, image enhancement and the normalized processing of pixel is carried out to obtain training example eye fundus image.
3. a kind of glaucoma image detecting method based on eye-ground photography deep learning according to claim 1, feature
It is:In the step 1, glaucoma and bottom of the normal eyes point that the training example eye fundus image in composing training database is carried
The quantitative proportion of class mark is 1:1.
4. a kind of glaucoma image detecting method based on eye-ground photography deep learning according to claim 1, feature
It is:The step 2 uses the method for mirror image and rotation and is expanded in the method that fixed dimension is divided in original image sliding window
Increase.
5. a kind of glaucoma image detecting method based on eye-ground photography deep learning according to claim 1, feature
It is:The step 2 is specially:
2.1) down-sampled to the progress of each Zhang Xunlian examples eye fundus image to obtain the eye fundus image of 256 × 256 resolution ratio, and to scheme
As the upper left corner be starting origin along horizontal axis to the right with establish sliding window under vertical axis, horizontal axis single direction to the right or under vertical axis
Sliding window is using the random integers between 0-32 as 4 of step-length sliding windows at equal intervals, image is further cut into 4 by sliding window ×
4=16 block subgraphs so that picture number is expanded to 4 × 4=16 times;
2.2) subgraph that step 2.1) obtains is rotated with random ratio with vertically and horizontally mirror image, 90 °, 180 ° and 270 °
Five kinds of methods carry out second and expand respectively, obtain training dataset.
6. a kind of glaucoma image detecting method based on eye-ground photography deep learning according to claim 1, feature
It is:When convolutional neural networks are trained, the Parameters of Neural Network Structure is adjusted according to the learning rate of setting, to obtain
The repeatedly convolutional neural networks after training.
7. a kind of glaucoma image detecting method based on eye-ground photography deep learning according to claim 1, feature
It is:The convolutional neural networks framework includes 14 layers of neuron, respectively be input layer, convolutional layer, maximum pond layer,
Convolutional layer, convolutional layer, convolutional layer, convolutional layer, maximum pond layer, full articulamentum, abandons layer, full articulamentum, loses maximum pond layer
Abandon layer and output layer;Wherein, five layers of convolutional layer C1~C5 are passed using the signal between ReLU activation primitives simulation biological neuron
It passing, preceding two layers of convolutional layer C1 and C2 are respectively adopted the convolution kernel that size is 11 × 11 and 5 × 5 and carry out convolution to input picture, after
Three-layer coil lamination C3~C5 then uses 3 × 3 convolution kernel;C1~C5 is convolutional layer and with ReLU (Rectified Linear
Units corrects linear unit) signal between activation primitive simulation biological neuron transmits, and C1 and C2 are respectively adopted size and are
11 × 11 and 5 × 5 convolution kernel carries out convolution to input picture, and C3~C5 is all made of 3 × 3 convolution kernel, and convolutional layer C1,
The window that C2, C5 are all made of 3 × 3 after activation primitive activates carries out maximum pond layer operation and by local acknowledgement's regularization
(Local Response Normalization, LRN) is connect with back layer grade;Wherein, two layers of full articulamentum F1 and F2 with
The form that neuron connects entirely is connected with front and back adjacent layer.
8. a kind of glaucoma image detecting method based on eye-ground photography deep learning according to claim 1, feature
It is:Convolutional neural networks use cross entropy cost function as loss function when training, and use the stochastic gradient based on momentum
Descent algorithm is as study optimization algorithm.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810360708.6A CN108665447B (en) | 2018-04-20 | 2018-04-20 | Glaucoma image detection method based on fundus photography deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810360708.6A CN108665447B (en) | 2018-04-20 | 2018-04-20 | Glaucoma image detection method based on fundus photography deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108665447A true CN108665447A (en) | 2018-10-16 |
CN108665447B CN108665447B (en) | 2021-07-30 |
Family
ID=63780209
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810360708.6A Active CN108665447B (en) | 2018-04-20 | 2018-04-20 | Glaucoma image detection method based on fundus photography deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108665447B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109215039A (en) * | 2018-11-09 | 2019-01-15 | 浙江大学常州工业技术研究院 | A kind of processing method of eyeground picture neural network based |
CN109528155A (en) * | 2018-11-19 | 2019-03-29 | 复旦大学附属眼耳鼻喉科医院 | A kind of intelligent screening system and its method for building up suitable for the concurrent open-angle glaucoma of high myopia |
CN109785308A (en) * | 2019-01-09 | 2019-05-21 | 上海联影医疗科技有限公司 | A kind of method, apparatus, equipment and storage medium that heart stroke solves |
CN109859172A (en) * | 2019-01-08 | 2019-06-07 | 浙江大学 | Based on the sugared net lesion of eyeground contrastographic picture deep learning without perfusion area recognition methods |
CN110120055A (en) * | 2019-04-12 | 2019-08-13 | 浙江大学 | Fundus fluorescein angiography image based on deep learning is without perfusion area automatic division method |
CN110432892A (en) * | 2019-08-05 | 2019-11-12 | 苏州米特希赛尔人工智能有限公司 | Machine learning ECG Automatic Diagnosis System |
CN110874595A (en) * | 2019-10-22 | 2020-03-10 | 杭州效准智能科技有限公司 | Multi-dish dinner plate intelligent segmentation method based on deep learning |
CN110969191A (en) * | 2019-11-07 | 2020-04-07 | 吉林大学 | Glaucoma prevalence probability prediction method based on similarity maintenance metric learning method |
CN111080630A (en) * | 2019-12-20 | 2020-04-28 | 腾讯医疗健康(深圳)有限公司 | Fundus image detection apparatus, method, device, and storage medium |
WO2020151307A1 (en) * | 2019-01-23 | 2020-07-30 | 平安科技(深圳)有限公司 | Automatic lesion recognition method and device, and computer-readable storage medium |
CN111476283A (en) * | 2020-03-31 | 2020-07-31 | 上海海事大学 | Glaucoma fundus image identification method based on transfer learning |
CN111784687A (en) * | 2020-07-22 | 2020-10-16 | 上海理工大学 | Glaucoma fundus image detection method based on deep learning |
CN111833311A (en) * | 2020-06-18 | 2020-10-27 | 安徽农业大学 | Image identification method based on deep learning and application of image identification method to rice disease identification |
CN113768460A (en) * | 2021-09-10 | 2021-12-10 | 北京鹰瞳科技发展股份有限公司 | Fundus image analysis system and method and electronic equipment |
CN114694236A (en) * | 2022-03-08 | 2022-07-01 | 浙江大学 | Eyeball motion segmentation positioning method based on cyclic residual convolution neural network |
CN116523903A (en) * | 2023-06-25 | 2023-08-01 | 天津师范大学 | Multi-mode fracture injury detection and identification method and system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102708370A (en) * | 2012-05-17 | 2012-10-03 | 北京交通大学 | Method and device for extracting multi-view angle image foreground target |
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 |
US9830529B2 (en) * | 2016-04-26 | 2017-11-28 | Xerox Corporation | End-to-end saliency mapping via probability distribution prediction |
-
2018
- 2018-04-20 CN CN201810360708.6A patent/CN108665447B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102708370A (en) * | 2012-05-17 | 2012-10-03 | 北京交通大学 | Method and device for extracting multi-view angle image foreground target |
US9830529B2 (en) * | 2016-04-26 | 2017-11-28 | Xerox Corporation | End-to-end saliency mapping via probability distribution prediction |
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 |
Non-Patent Citations (2)
Title |
---|
XIANGYU CHEN ETAL.: "Glaucoma detection based on deep convolutional neural network", 《2015 37TH ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY (EMBC)》 * |
陈才扣等: "基于泛滑动窗与加权2DLDA的单样本人脸识别", 《计算机工程与应用》 * |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109215039A (en) * | 2018-11-09 | 2019-01-15 | 浙江大学常州工业技术研究院 | A kind of processing method of eyeground picture neural network based |
CN109528155A (en) * | 2018-11-19 | 2019-03-29 | 复旦大学附属眼耳鼻喉科医院 | A kind of intelligent screening system and its method for building up suitable for the concurrent open-angle glaucoma of high myopia |
CN109859172A (en) * | 2019-01-08 | 2019-06-07 | 浙江大学 | Based on the sugared net lesion of eyeground contrastographic picture deep learning without perfusion area recognition methods |
CN109785308A (en) * | 2019-01-09 | 2019-05-21 | 上海联影医疗科技有限公司 | A kind of method, apparatus, equipment and storage medium that heart stroke solves |
WO2020151307A1 (en) * | 2019-01-23 | 2020-07-30 | 平安科技(深圳)有限公司 | Automatic lesion recognition method and device, and computer-readable storage medium |
CN110120055A (en) * | 2019-04-12 | 2019-08-13 | 浙江大学 | Fundus fluorescein angiography image based on deep learning is without perfusion area automatic division method |
CN110432892A (en) * | 2019-08-05 | 2019-11-12 | 苏州米特希赛尔人工智能有限公司 | Machine learning ECG Automatic Diagnosis System |
CN110874595A (en) * | 2019-10-22 | 2020-03-10 | 杭州效准智能科技有限公司 | Multi-dish dinner plate intelligent segmentation method based on deep learning |
CN110969191A (en) * | 2019-11-07 | 2020-04-07 | 吉林大学 | Glaucoma prevalence probability prediction method based on similarity maintenance metric learning method |
CN111080630A (en) * | 2019-12-20 | 2020-04-28 | 腾讯医疗健康(深圳)有限公司 | Fundus image detection apparatus, method, device, and storage medium |
CN111080630B (en) * | 2019-12-20 | 2024-03-08 | 腾讯医疗健康(深圳)有限公司 | Fundus image detection device, fundus image detection method, fundus image detection device, and fundus image storage medium |
CN111476283A (en) * | 2020-03-31 | 2020-07-31 | 上海海事大学 | Glaucoma fundus image identification method based on transfer learning |
CN111833311A (en) * | 2020-06-18 | 2020-10-27 | 安徽农业大学 | Image identification method based on deep learning and application of image identification method to rice disease identification |
CN111833311B (en) * | 2020-06-18 | 2023-12-22 | 安徽农业大学 | Image recognition method based on deep learning and application of image recognition method in rice disease recognition |
CN111784687A (en) * | 2020-07-22 | 2020-10-16 | 上海理工大学 | Glaucoma fundus image detection method based on deep learning |
CN113768460A (en) * | 2021-09-10 | 2021-12-10 | 北京鹰瞳科技发展股份有限公司 | Fundus image analysis system and method and electronic equipment |
CN113768460B (en) * | 2021-09-10 | 2023-11-14 | 北京鹰瞳科技发展股份有限公司 | Fundus image analysis system, fundus image analysis method and electronic equipment |
CN114694236A (en) * | 2022-03-08 | 2022-07-01 | 浙江大学 | Eyeball motion segmentation positioning method based on cyclic residual convolution neural network |
CN116523903A (en) * | 2023-06-25 | 2023-08-01 | 天津师范大学 | Multi-mode fracture injury detection and identification method and system |
Also Published As
Publication number | Publication date |
---|---|
CN108665447B (en) | 2021-07-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108665447A (en) | A kind of glaucoma image detecting method based on eye-ground photography deep learning | |
Chen et al. | Automatic feature learning for glaucoma detection based on deep learning | |
CN109859172A (en) | Based on the sugared net lesion of eyeground contrastographic picture deep learning without perfusion area recognition methods | |
Sertkaya et al. | Diagnosis of eye retinal diseases based on convolutional neural networks using optical coherence images | |
CN108537282A (en) | A kind of diabetic retinopathy stage division using extra lightweight SqueezeNet networks | |
CN108921227A (en) | A kind of glaucoma medical image classification method based on capsule theory | |
CN109691979A (en) | A kind of diabetic retina image lesion classification method based on deep learning | |
CN110120055A (en) | Fundus fluorescein angiography image based on deep learning is without perfusion area automatic division method | |
CN112446860B (en) | Automatic screening method for diabetic macular edema based on transfer learning | |
CN112580580A (en) | Pathological myopia identification method based on data enhancement and model fusion | |
Afroze et al. | Glaucoma detection using inception convolutional neural network v3 | |
Gupta et al. | An efficient model for detection and classification of internal eye diseases using deep learning | |
CN111028230A (en) | Fundus image optic disc and macula lutea positioning detection algorithm based on YOLO-V3 | |
Yadav et al. | Computer‐aided diagnosis of cataract severity using retinal fundus images and deep learning | |
Pandey et al. | Detection of glaucoma disease using image processing, soft computing and deep learning approaches | |
Niya et al. | Analysis of different automatic cataract detection and classification methods | |
Shaban et al. | Low-complexity computer-aided diagnosis for diabetic retinopathy | |
CN111784687A (en) | Glaucoma fundus image detection method based on deep learning | |
Mahendran et al. | Morphological process based segmentation for the detection of exudates from the retinal images of diabetic patients | |
Suedumrong et al. | Application of deep convolutional neural networks vgg-16 and googlenet for level diabetic retinopathy detection | |
Liu et al. | Retinal vessel segmentation using densely connected convolution neural network with colorful fundus images | |
Aydin et al. | Domain modelling for a lightweight convolutional network focused on automated exudate detection in retinal fundus images | |
Purwanithami et al. | Hemorrhage diabetic retinopathy detection based on fundus image using neural network and FCM segmentation | |
Munarto et al. | Automatic Cataract Classification System Using Neural Network Algorithm Backpropagation | |
Hussein et al. | Convolutional Neural Network in Classifying Three Stages of Age-Related Macula Degeneration |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |