CN108021916B - Deep learning diabetic retinopathy sorting technique based on attention mechanism - Google Patents
Deep learning diabetic retinopathy sorting technique based on attention mechanism Download PDFInfo
- Publication number
- CN108021916B CN108021916B CN201711497342.9A CN201711497342A CN108021916B CN 108021916 B CN108021916 B CN 108021916B CN 201711497342 A CN201711497342 A CN 201711497342A CN 108021916 B CN108021916 B CN 108021916B
- Authority
- CN
- China
- Prior art keywords
- network
- attention
- neural network
- diabetic retinopathy
- parameter
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color 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 present invention discloses a kind of deep learning diabetic retinopathy sorting technique based on attention mechanism, including：A series of eye fundus images are chosen as primary data sample, are divided into training set and test set after pretreatment, cutting is normalized to sample；Parameter initialization and fine tuning are carried out to main neural network, training set image is input to main neural network and is trained, generates characteristic pattern；The parameter of fixed main neural network trains attention network using training set image, exports lesion candidate regions degree figure and normalizes to gain attention and tries hard to, and will pay attention to trying hard to gaining attention power mechanism with characteristic pattern product；Attention mechanism is acquired into result and inputs main neural network, continues to train using training set image, finally obtains diabetic retinopathy grade separation model.By introducing attention mechanism, using diabetic retinopathy area data set pair, it is trained the present invention, enhances the information characteristics of lesion region while retaining network primitive character.
Description
Technical field
The present invention relates to the deep learning diabetic retinopathy sorting techniques based on attention mechanism, belong to medicine figure
As process field.
Background technology
Clinically doctor is by observing and analyzing retina eyeground figure for the diagnosis of diabetic retinopathy at present
It is carried out as the upper early stage pathology feature such as symptoms such as aneurysms, hard exudate and bleeding.In practice, diabetes view
Film lesions type is more, and lesion is various, sufferer severity differs, and causes oculist's difficult diagnosis.Therefore, extensive
Diabetic retinopathy disorder in screening in, computer-aided diagnosis technology can mitigate the burden of doctor significantly, and quickly,
Effectively auxiliary doctor realizes classification diagnosis.
In the automatic diagnosis algorithm of Most current, biography is based primarily upon for the classification of diabetic retinopathy eye fundus image
System manual method designs extraction feature, then carries out the structure of grader.Such as using including shape, color, brightness and priori
The craft feature such as knowledge carries out diabetic retinopathy detection, these methods can only can obtain preferably on small data set
As a result, since manual features extraction process is cumbersome, efficiency is low in the case of large data sets and poor robustness.With artificial intelligence
The development of algorithm has had researcher to propose the diabetic retinopathy classification diagnosis side for being directly based upon deep learning at present
Method, such as convolutional neural networks are directly connected to the classification task that eye fundus image carries out diabetic retinopathy, such method
It is designed for all diabetic retinopathy types, only regards convolutional neural networks as a flight data recorder, not
Have and the lesion distributed intelligence closely related with diagnosis is taken into account, lacks effectively and intuitively explain.
Invention content
Goal of the invention：Present invention aims in view of the deficiencies of the prior art, provide a kind of depth based on attention mechanism
Degree study diabetic retinopathy sorting technique, this method introduce attention mechanism in depth convolutional network, will pay attention to
In power internet startup disk to depth network, and it is instructed using the diabetic retinopathy area data set pair of expert's mark
Practice, attention network can introduce expertise, generate the lesion area-of-interest for including candidate lesion region position, this method
The information characteristics of lesion region can be enhanced while retaining network primitive character.
Technical solution：Deep learning diabetic retinopathy sorting technique of the present invention based on attention mechanism,
It is characterised in that it includes following steps：
(1) a series of eyeground figures in EyePACS data sets, DiaretDB1 data sets, Messidor data sets are chosen respectively
Picture is used as primary data sample, and pretreatment is normalized to eye fundus image, and cutting is carried out after pretreatment and ensures that size is identical, will
Eye fundus image after cutting is divided into training set and test set；
(2) convolutional neural networks model is built, the convolutional neural networks model includes main neural network and attention net
Network；Individually main neural network is trained using ImageNet parameters, by the obtained parameter of training to main neural network into
Row finely tunes and preserves main neural network model parameter；In the main neural network model parameter of preservation, diabetic retina is chosen
The best main neural network model parameter of lesion grade separation initializes the main neural network parameter portion in convolutional neural networks
Point, remaining stochastic parameter initialization；
(3) the training set image in EyePACS data sets is input to main neural network to be trained, generates characteristic pattern；
The parameter of fixed main neural network trains attention network, attention net using the training set image in DiaretDB1 data sets
Network exports a lesion candidate region gray-scale map；
(4) the lesion candidate region gray-scale map that attention network generates is normalized to gain attention and is tried hard to, and will note
The characteristic pattern trying hard to export with main neural network anticipate into row element dot product, product gains attention power mechanism；
(5) attention mechanism is acquired result to input in main neural network, using the training set figure in EyePACS data sets
As continuing to train, the parameter of main neural network is adjusted when training according to the learning rate of setting, diabetes view may finally be obtained
Film lesion grade separation model.
The training set of the classification of DR 5 is carried out with EyePACs data sets, DiaretDB1 data sets are the DR lesions of expert's mark
The data set in region is the data set of another DR classification for training attention network portion, Messidor data sets, is used for
Verify the robustness of network.
Above-mentioned technical proposal is further improved, the pretreatment operation of the step (1) is：Scheme in extraction primary data sample
The foreground area of picture, using following formula to pretreatment is normalized,
Ic(x, y)=α I (x, y)+β Gaussion (x, y, ρ) * I (x, y)+γ
Wherein, I is input picture, and * indicates that convolution operation, Gaussion (x, y, ρ) indicate that standard deviation is the gaussian filtering of ρ
Device, parameter alpha, beta, gamma and ρ are rule of thumb set as α=4, β=- 4, using eye fundus image center as the center of circle, erode to eyeground figure side
The region of edge 5%；
It is 720 × 720 by pretreated image cropping, and the image for being divided into training set is subjected to Random-Rotation
0 °/90 °/180 °/270 ° to realize image data augmentation.
Further, using the method for transfer learning in the step (2), the parameter pair that training obtains on ImageNet
Main neural network is finely adjusted, and selects the best main neural network model parameter of diabetic retinopathy grade separation to convolution
Main neural network is initialized in neural network, and random initializtion remainder parameter.
Further, it will be carried out after the output normalization of attention network in the step (4) plus 1 operates, acquired results
Dot product operation is carried out with the characteristic pattern of main neural network certain layer, is then input in main neural network and carries out subsequent training.
Characteristic layer is conv2d_2b_3x3 convolutional layers, and characteristic pattern is Feature map M.
Further, the attention network is a symmetrical full convolutional network, includes 5 convolution of down-sampling process
5 uncoiling laminations of layer and upsampling process.
Further, the main neural network uses inception-resnet-v2, integrates residual error learning structure and perception
Two kinds of modules of structure.
Optimization in the step (3) to (5) using Nesterov Momentum algorithms as convolutional neural networks is calculated
Method, the momentum Optimization Factor used are 0.9 all weights of update；In the repetitive exercise each time of network, using mean square error
Loss function as diabetic retinopathy grade separation trains neural networks at different levels, in error amount back-propagation process
In, Grad, and the parameter of calculated Grad update network are calculated using Nesterov Momentum algorithms,
It is 0.0005 to use L2 regularization terms, weight decay factor to all parameters of network；According to the loss function of attention network,
Grad is calculated using Nesterov Momentum algorithms, the parameter of update attention network completes the iteration mistake successively of network
Journey.
The present invention makees optimization algorithm using Nesterov Momentum algorithms (being one kind of stochastic gradient algorithm), simultaneously
Neural networks at different levels are trained as loss function using mean square error (Mean Squared Error, MSD), to parameter in network
Using L2Weight Decay regularization, and batch-normlization methods are used in a network.Optimization algorithm is applied to
In the training process of network, it is used for the update of network weight, gradient is calculated during the error back propagation of network, updates net
Network parameter.
Advantageous effect：1, attention mechanism is introduced in depth convolutional network, by attention internet startup disk to depth network
In, and it is trained using the diabetic retinopathy area data set pair of expert's mark, attention network can introduce
Expertise generates the lesion area-of-interest for including candidate lesion region position.
2, it using the depth network based on residual error perceptual structure, and will be noted using feature Enhancement Method in the training of network
The lesion candidate region knowledge that meaning power network generates is introduced into diabetic retinopathy classification task.This method is retaining original
The corresponding characteristic information of area-of-interest is enhanced on the basis of beginning characteristic information.
3, in network model diagnosis of diabetic retinopathy result is improved using the loss function based on mean square error.
The model is a kind of multi-task learning model, simultaneously can be to the lesion of eye fundus image promoting classification performance using expertise
Area carries out coarse positioning, has good robustness.
Description of the drawings
Fig. 1：The implementation process frame diagram of the present invention
Fig. 2：Attention function structure chart
Fig. 3：The front and back comparison diagram of eye fundus image normalization
Fig. 4：The lesion candidate region that attention structural network generates and practical expert's tab area comparison diagram.
Specific implementation mode
Technical solution of the present invention is described in detail below by attached drawing, but protection scope of the present invention is not limited to
The embodiment.
Embodiment 1：Deep learning diabetic retinopathy grade separation provided by the invention based on attention mechanism
Method is detected identification to diabetic retinopathy grade, and concrete operations carry out as follows：
1, data set is chosen；
(1) EyePACS data sets
Five categorized data set of diabetic retinopathy disease grade includes 88702 coloured silks from 44315 patients
Color eye fundus image, the resolution ratio of image is between.The data set is divided into two parts：Training set 35126, which is opened, (comes from 17563
Patient), test set 53576 is opened and (comes from 26788 patients).The DR menace levels of every eye fundus image are by doctor according to ETDRS
Table is labeled：' 0 ' indicates no diabetic retinopathy, and ' 1 ' indicates slight non-appreciation diabetic retinopathy,
' 2 ' indicate medium diabetes mellitus retinopathy, and ' 3 ' indicate that severe diabetic retinopathy, ' 4 ' expression appreciation diabetes regard
Retinopathy.EyePACS data sets have following features：(1) as shown in table 1, the category distribution of the data set is seriously uneven,
The data accounting of middle classification 0 is 79%, and the data accounting of classification 3 and 4 is respectively 2% and 2%；(2) adopting due to the data set
For collection from different fundus cameras, the resolution ratio and picture quality of image have larger otherness.
Table 1EyePACS data distributions
(2) DiaretDB1 data sets
The data set shares 89 colored eye fundus images, and every eye fundus image is marked out potential manually by 4 medical experts
Focal area, including aneurysms, hard exudate, bleeding and cotton-wool spot these lesion regions.The data are concentrated with 5
Picture is no diabetic retinopathy, and in addition in 84 images, a kind of lesion region is included at least in base map of often opening one's eyes.Mark
The label image of note saves as the gray level image of 0-255, and wherein bigger expression region lesion of gray value is more serious.
(3) Messidor data sets
The data set shares 1200 colored eye fundus images, wherein 540 normal pictures 660 open illness image, including three
Kind resolution ratio：1440x960,2240x1488 and 2304x1536, image are TIF formats.It is oozed according to microaneurysm, bleeding, hardness
Go out DR lesion grade of the lesions such as object to every image labeling from 0 to 3, wherein ' 0 ' image of label there are 546 to account for total amount of data
46%, ' 1 ' image of label has 154 to account for the 12.75% of total amount of data, and ' 2 ' image of label has 247 to account for total amount of data
20.58%, ' 3 ' image of label has 253 to account for the 21.67% of total amount of data.
2, data prediction
It is overseas for the useful region of interest of checkout and diagnosis due in eye fundus image, removing, it is further comprised in image big
The background area of amount, therefore before being diagnosed, we extract foreground area to content of interest first.
Ic(x, y)=α I (x, y)+β Gaussion (x, y, ρ) * I (x, y)+γ
Eye fundus image normalization is carried out according to above formula, obtains pretreated normalized image.Wherein, I is input picture, *
Indicate that convolution operation, Gaussion (x, y, ρ) indicate that standard deviation is the Gaussian filter of ρ, parameter alpha, beta, gamma and ρ are rule of thumb set
It is set to α=4, β=- 4.Since in the visual field, the marginal portion of eyeground figure usually will appear halation phenomenon, it is with eye fundus image center
The center of circle erodes to the region at eyeground figure edge 5%.Eye fundus image is as shown in Figure 3 after being normalized.
In this method, the input image size used is chosen for 720 × 720, and passes through Random-Rotation on training set
(0°/90°/180°/270°) and overturn at random to carrying out data augmentation by pretreated data set.
3, training convolutional neural networks model
In network training process, data are input to network in batches, and a lot data includes two queues：DiaretDB1
Data set and EyePACS data sets, in the repetitive exercise each time of network：DiaretDB1 data sets are sent into convolutional Neural net
Network generates lesion candidate region gray-scale map by attention network, the error of attention network is calculated by MSE loss functions；
EyePACS data are sent into convolutional neural networks, and corresponding lesion candidate region information, attention net are generated by attention network
It is carried out after the output normalization of network plus 1 operates, the characteristic pattern of acquired results and main neural network certain layer conv2d_2b_3x3
Feature Map M carry out dot product operation, are then input in main neural network and carry out subsequent training；Utilize mean square error
(mean-square error, MSE) is as calculating diabetic retinopathy (diabetic retinopathy, DR) grade
The loss function of classification trains neural networks at different levels.According to attention network losses function, the parameter of attention network is updated；Root
According to DR grade separation loss functions, main neural network corresponding part parameter is updated, completes the iterative process successively of network.
The lesion region candidate that DiaretDB1 data sets generate is schemed to melt with clinical information by the training of attention network
It closes, this method can be introduced into expertise in network and improve classification performance.Fig. 4 is the lesion that attention structural network generates
The comparative examples of candidate region and practical expert's tab area.
In the training process, the present invention uses the stochastic gradient descent method (Stochastic based on mini-batch
Gradient Descent, SGD) it optimizes, the SGD for the use of momentum Optimization Factor being 0.9 updates all weights.Network uses
Based on MSE as loss function, and it is 0.0005 to use L2 regularization terms, weight decay factor to each parameter in network.
4, handling result is analyzed
This method carrys out quantification treatment as a result, being respectively using following four performance metric：Classification accuracy (accuracy,
ACC), specific (specificity, SPE), sensibility (sensitivity, SEN) and AUC (the area under ROC
Curve) value.For classification results, addition is used based on secondary weighted kappa values as another performance metric, it can be with
Weigh predicted value and the direct consistency of physical tags.Kappa value calculations are as follows：
To the confusion matrix O, O of one N × N of image prediction label configurationsi,jThe amount of images of (i, j) is designated as under corresponding to,
Weighting matrix is expressed as：
EN×NIndicate the confusion matrix of image true tag construction, it is assumed that without correlation between prediction label and true tag
Property, then secondary weighted kappa values are as follows：
Handling result of this method on EyePACS data sets used possesses 0.840 kappa values on verification collection, surveys
Possess 0.835 kappa values on examination collection.Our method is further verified on Messidor data sets.We
Method accuracy, AUC value indices on all achieve fabulous result.Specifically, this method is in referable/non-
Achieve 91.6% accuracy rate in referable tasks, 90.3% sensibility, 95.2% specificity and 0.963
AUC value.
It uses the eye fundus image more than 30,000 by professional's mark in the present embodiment to be trained, 50,000
82% accuracy rate is reached in the test data set of multiple.Many experiments show proposed by the present invention based on attention mechanism
Deep learning method have higher classification performance.Using the deep learning diabetic retina above based on attention mechanism
The disaggregated model that lesion classification method is established can carry out automation classification to diabetic retinopathy, and in category distribution
There is good robustness, this has having very important significance on medical domain in unbalanced data.
As described above, although the present invention has been indicated and described with reference to specific preferred embodiment, must not explain
For the limitation to invention itself.It without prejudice to the spirit and scope of the invention as defined in the appended claims, can be right
Various changes can be made in the form and details for it.
Claims (7)
1. the deep learning diabetic retinopathy sorting technique based on attention mechanism, which is characterized in that including walking as follows
Suddenly：
(1) a series of eye fundus images in EyePACS data sets, DiaretDB1 data sets, Messidor data sets are chosen respectively to make
For primary data sample, pretreatment is normalized to eye fundus image, cutting is carried out after pretreatment and ensures that size is identical, will be cut
Eye fundus image afterwards is divided into training set and test set；
(2) convolutional neural networks model is built, the convolutional neural networks model includes main neural network and attention network；It adopts
Individually main neural network is trained with ImageNet data sets, the parameter obtained by training carries out main neural network micro-
It adjusts and preserves main neural network model parameter；In the main neural network model parameter of preservation, diabetic retinopathy is chosen
The best main neural network model parameter of grade separation initializes the main neural network parameter part in convolutional neural networks, remaining
Lower stochastic parameter initialization；
(3) the training set image in EyePACS data sets is input to main neural network to be trained, generates characteristic pattern；It is fixed
The parameter of main neural network trains attention network, attention network defeated using the training set image in DiaretDB1 data sets
Go out lesion candidate region gray-scale map；
(4) the lesion candidate region gray-scale map that attention network generates is normalized to gain attention and is tried hard to, and by attention
Into row element dot product, product gains attention power mechanism the characteristic pattern of figure and the output of main neural network；
(5) attention mechanism result is acquired to input in main neural network, using the training set image in EyePACS data sets after
Continuous training, the parameter of main neural network is adjusted according to the learning rate of setting, finally obtains diabetic retinopathy etc. when training
Grade disaggregated model.
2. the deep learning diabetic retinopathy sorting technique according to claim 1 based on attention mechanism,
It is characterized in that：The pretreatment operation of the step (1) is：The foreground area for extracting image in primary data sample, using following formula
To pretreatment is normalized,
Ic(x, y)=α I (x, y)+β Gaussion (x, y, ρ) * I (x, y)+γ
Wherein, I is input picture, and * indicates that convolution operation, Gaussion (x, y, ρ) indicate that standard deviation is the Gaussian filter of ρ,
Parameter alpha, beta, gamma and ρ are rule of thumb set as α=4, β=- 4, using eye fundus image center as the center of circle, erode to eyeground figure edge
5% region；
By pretreated image cropping be 720 × 720, and by be divided into training set image carry out 0 ° of Random-Rotation/
90 °/180 °/270 ° to realize image data augmentation.
3. the deep learning diabetic retinopathy sorting technique according to claim 1 based on attention mechanism,
It is characterized in that：Using the method for transfer learning in the step (2), the parameter that training obtains on ImageNet to main nerve net
Network is finely adjusted, and selects the best main neural network model parameter of diabetic retinopathy grade separation to convolutional neural networks
Middle main neural network is initialized, and random initializtion remainder parameter.
4. the deep learning diabetic retinopathy sorting technique according to claim 1 based on attention mechanism,
It is characterized in that：It will be carried out after the output normalization of attention network in the step (4) plus 1 operates, acquired results and main nerve
The characteristic pattern Feature Map M of network certain layer conv2d_2b_3x3 carry out dot product operation, are then input to main neural network
It is middle to carry out subsequent training.
5. the deep learning diabetic retinopathy sorting technique according to claim 1 based on attention mechanism,
It is characterized in that：The attention network is a symmetrical full convolutional network, including 5 convolutional layers of down-sampling process and is above adopted
5 uncoiling laminations of sample process.
6. the deep learning diabetic retinopathy sorting technique according to claim 1 based on attention mechanism,
It is characterized in that：The main neural network uses inception-resnet-v2, integrates two kinds of residual error learning structure and perceptual structure
Module.
7. the deep learning diabetic retinopathy sorting technique according to claim 1 based on attention mechanism,
It is characterized in that：Optimization in the step (3) to (5) using Nesterov Momentum algorithms as convolutional neural networks is calculated
Method, the momentum Optimization Factor used are 0.9 all weights of update；In the repetitive exercise each time of network, using mean square error
Loss function as diabetic retinopathy grade separation trains neural networks at different levels, in error amount back-propagation process
In, Grad, and the parameter of calculated Grad update network are calculated using Nesterov Momentum algorithms,
It is 0.0005 to use L2 regularization terms, weight decay factor to all parameters of network；According to the loss function of attention network,
Grad is calculated using Nesterov Momentum algorithms, the parameter of update attention network completes the iteration mistake successively of network
Journey.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711497342.9A CN108021916B (en) | 2017-12-31 | 2017-12-31 | Deep learning diabetic retinopathy sorting technique based on attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711497342.9A CN108021916B (en) | 2017-12-31 | 2017-12-31 | Deep learning diabetic retinopathy sorting technique based on attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108021916A CN108021916A (en) | 2018-05-11 |
| CN108021916B true CN108021916B (en) | 2018-11-06 |
Family
ID=62071224
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711497342.9A Active CN108021916B (en) | 2017-12-31 | 2017-12-31 | Deep learning diabetic retinopathy sorting technique based on attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108021916B (en) |
Families Citing this family (83)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108734290B (en) * | 2018-05-16 | 2021-05-18 | 湖北工业大学 | Convolutional neural network construction method based on attention mechanism and application |
| CN108734138B (en) * | 2018-05-24 | 2020-10-30 | 浙江工业大学 | Melanoma skin disease image classification method based on ensemble learning |
| CN110648303B (en) * | 2018-06-08 | 2022-07-26 | 上海市第六人民医院 | Fundus image analysis method, computer device, and storage medium |
| CN108876775B (en) * | 2018-06-12 | 2022-10-18 | 湖南大学 | Method for rapidly detecting diabetic retinopathy |
| CN108765422A (en) * | 2018-06-13 | 2018-11-06 | 云南大学 | A kind of retinal images blood vessel automatic division method |
| CN108876776B (en) * | 2018-06-13 | 2021-08-24 | 东软集团股份有限公司 | Classification model generation method, fundus image classification method and device |
| CN109034196A (en) * | 2018-06-21 | 2018-12-18 | 北京健康有益科技有限公司 | Model generating method and device, food recognition methods and device |
| CN108898175B (en) * | 2018-06-26 | 2020-11-20 | 北京工业大学 | Computer-aided model construction method based on deep learning gastric cancer pathological section |
| CN109325398B (en) * | 2018-06-30 | 2020-10-09 | 东南大学 | Human face attribute analysis method based on transfer learning |
| CN109003264B (en) * | 2018-07-05 | 2022-05-06 | 腾讯医疗健康(深圳)有限公司 | Retinopathy image type identification method and device and storage medium |
| CN108960257A (en) * | 2018-07-06 | 2018-12-07 | 东北大学 | A kind of diabetic retinopathy grade stage division based on deep learning |
| CN108960260B (en) * | 2018-07-12 | 2020-12-29 | 东软集团股份有限公司 | Classification model generation method, medical image classification method and medical image classification device |
| CN109102502B (en) * | 2018-08-03 | 2021-07-23 | 西北工业大学 | Pulmonary nodule detection method based on three-dimensional convolutional neural network |
| CN109376767B (en) * | 2018-09-20 | 2021-07-13 | 中国科学技术大学 | Retina OCT image classification method based on deep learning |
| CN110929836B (en) * | 2018-09-20 | 2023-10-31 | 北京市商汤科技开发有限公司 | Neural network training and image processing method and device, electronic equipment and medium |
| CN109543529A (en) * | 2018-10-19 | 2019-03-29 | 北京陌上花科技有限公司 | Biopsy method and device |
| CN109685819B (en) * | 2018-12-11 | 2021-02-26 | 厦门大学 | Three-dimensional medical image segmentation method based on feature enhancement |
| US10963757B2 (en) | 2018-12-14 | 2021-03-30 | Industrial Technology Research Institute | Neural network model fusion method and electronic device using the same |
| CN109344920B (en) * | 2018-12-14 | 2021-02-02 | 汇纳科技股份有限公司 | Customer attribute prediction method, storage medium, system and device |
| CN109754404B (en) * | 2019-01-02 | 2020-09-01 | 清华大学深圳研究生院 | End-to-end tumor segmentation method based on multi-attention mechanism |
| CN109902717A (en) * | 2019-01-23 | 2019-06-18 | 平安科技（深圳）有限公司 | Lesion automatic identifying method, device and computer readable storage medium |
| CN109858429B (en) * | 2019-01-28 | 2021-01-19 | 北京航空航天大学 | Eye ground image lesion degree identification and visualization system based on convolutional neural network |
| CN111507932B (en) * | 2019-01-31 | 2023-05-09 | 福州依影健康科技有限公司 | High-specificity diabetic retinopathy characteristic detection method and storage device |
| CN109919831B (en) * | 2019-02-13 | 2023-08-25 | 广州视源电子科技股份有限公司 | Method, electronic device and computer readable storage medium for migrating retinal fundus images in different image domains |
| CN109919915B (en) * | 2019-02-18 | 2021-03-23 | 广州视源电子科技股份有限公司 | Retina fundus image abnormal region detection method and device based on deep learning |
| CN111767929A (en) * | 2019-03-14 | 2020-10-13 | 上海市第一人民医院 | Method and system for constructing sub-macular neovascularization model |
| CN109934823A (en) * | 2019-03-25 | 2019-06-25 | 天津工业大学 | A kind of DR eye fundus image macular edema stage division based on deep learning |
| CN109948719B (en) * | 2019-03-26 | 2023-04-18 | 天津工业大学 | Automatic fundus image quality classification method based on residual dense module network structure |
| CN109949302A (en) * | 2019-03-27 | 2019-06-28 | 天津工业大学 | Retinal feature Structural Techniques based on pixel |
| CN111753862A (en) * | 2019-03-29 | 2020-10-09 | 北京地平线机器人技术研发有限公司 | Method and device for training neural network model and image recognition method |
| CN110136096A (en) * | 2019-04-02 | 2019-08-16 | 成都真实维度科技有限公司 | A method of lesion region segmentation is carried out based on faulted scanning pattern data set |
| CN110097559B (en) * | 2019-04-29 | 2024-02-23 | 李洪刚 | Fundus image focus region labeling method based on deep learning |
| CN110084252B (en) * | 2019-04-29 | 2023-09-29 | 上海科锐克医药科技有限公司 | Deep learning-based diabetic retinopathy image labeling method |
| CN110097545A (en) * | 2019-04-29 | 2019-08-06 | 南京星程智能科技有限公司 | Eye fundus image generation method based on deep learning |
| CN110084803B (en) * | 2019-04-29 | 2024-02-23 | 靖松 | Fundus image quality evaluation method based on human visual system |
| CN111932486A (en) * | 2019-05-13 | 2020-11-13 | 四川大学 | Brain glioma segmentation method based on 3D convolutional neural network |
| CN110189334B (en) * | 2019-05-28 | 2022-08-09 | 南京邮电大学 | Medical image segmentation method of residual error type full convolution neural network based on attention mechanism |
| CN110211685B (en) * | 2019-06-10 | 2020-08-28 | 珠海上工医信科技有限公司 | Sugar network screening network structure model based on complete attention mechanism |
| CN110210570A (en) * | 2019-06-10 | 2019-09-06 | 上海延华大数据科技有限公司 | The more classification methods of diabetic retinopathy image based on deep learning |
| CN110236483B (en) * | 2019-06-17 | 2021-09-28 | 杭州电子科技大学 | Method for detecting diabetic retinopathy based on depth residual error network |
| CN110223291B (en) * | 2019-06-20 | 2021-03-19 | 南开大学 | Network method for training fundus lesion point segmentation based on loss function |
| CN110459303B (en) * | 2019-06-27 | 2022-03-08 | 浙江工业大学 | Medical image abnormity detection device based on depth migration |
| TWI702615B (en) * | 2019-07-26 | 2020-08-21 | 長佳智能股份有限公司 | Retinopathy assessment model establishment method and system |
| CN110598582A (en) * | 2019-08-26 | 2019-12-20 | 深圳大学 | Eye image processing model construction method and device |
| CN110796166B (en) * | 2019-09-25 | 2022-07-26 | 浙江大学 | Attention mechanism-based multitask image processing method |
| CN110728312B (en) * | 2019-09-29 | 2022-04-29 | 浙江大学 | Dry eye grading system based on regional self-adaptive attention network |
| CN110796249B (en) * | 2019-09-29 | 2024-01-05 | 中山大学孙逸仙纪念医院 | Ear endoscope image neural network model construction method and classification processing method |
| CN110458249B (en) * | 2019-10-10 | 2020-01-07 | 点内（上海）生物科技有限公司 | Focus classification system based on deep learning and probabilistic imaging omics |
| CN110689089A (en) * | 2019-10-12 | 2020-01-14 | 电子科技大学 | Active incremental training method for deep learning of multi-class medical image classification |
| CN110837803B (en) * | 2019-11-07 | 2022-11-29 | 复旦大学 | Diabetic retinopathy grading method based on depth map network |
| CN110969191B (en) * | 2019-11-07 | 2022-10-25 | 吉林大学 | Glaucoma prevalence probability prediction method based on similarity maintenance metric learning method |
| CN115049602A (en) * | 2019-11-28 | 2022-09-13 | 深圳硅基智控科技有限公司 | Optimization method of artificial neural network module |
| CN110929807B (en) * | 2019-12-06 | 2021-04-06 | 腾讯科技（深圳）有限公司 | Training method of image classification model, and image classification method and device |
| CN110992364B (en) * | 2019-12-31 | 2023-11-28 | 重庆艾可立安医疗器械有限公司 | Retina image recognition method, retina image recognition device, computer equipment and storage medium |
| CN111242168B (en) * | 2019-12-31 | 2023-07-21 | 浙江工业大学 | Human skin image lesion classification method based on multi-scale attention features |
| CN111275714B (en) * | 2020-01-13 | 2022-02-01 | 武汉大学 | Prostate MR image segmentation method based on attention mechanism 3D convolutional neural network |
| CN111259982B (en) * | 2020-02-13 | 2023-05-12 | 苏州大学 | Attention mechanism-based premature infant retina image classification method and device |
| CN111445440B (en) * | 2020-02-20 | 2023-10-31 | 上海联影智能医疗科技有限公司 | Medical image analysis method, device and storage medium |
| CN111369506B (en) * | 2020-02-26 | 2022-08-02 | 四川大学 | Lens turbidity grading method based on eye B-ultrasonic image |
| CN111507952A (en) * | 2020-04-13 | 2020-08-07 | 上海泗科智能科技有限公司 | Embedded-end diabetic retinopathy screening solution |
| KR102438659B1 (en) * | 2020-04-14 | 2022-08-30 | 충북대학교 산학협력단 | The method for classifying diabetic macular edema and the device thereof |
| CN111476312B (en) * | 2020-04-24 | 2022-04-19 | 南京图格医疗科技有限公司 | Method for classifying lesion images based on convolutional neural network |
| CN111816308B (en) * | 2020-07-13 | 2023-09-29 | 中国医学科学院阜外医院 | System for predicting coronary heart disease onset risk through facial image analysis |
| CN111882551B (en) * | 2020-07-31 | 2024-04-05 | 北京小白世纪网络科技有限公司 | Pathological image cell counting method, system and device |
| CN112101424B (en) * | 2020-08-24 | 2023-08-04 | 深圳大学 | Method, device and equipment for generating retinopathy identification model |
| CN112016626B (en) * | 2020-08-31 | 2023-12-01 | 中科泰明(南京)科技有限公司 | Uncertainty-based diabetic retinopathy classification system |
| CN112215239A (en) * | 2020-09-15 | 2021-01-12 | 浙江工业大学 | Retinal lesion fine-grained grading method and device |
| CN111938569A (en) * | 2020-09-17 | 2020-11-17 | 南京航空航天大学 | Eye ground multi-disease classification detection method based on deep learning |
| CN112185523B (en) * | 2020-09-30 | 2023-09-08 | 南京大学 | Diabetic retinopathy classification method based on multi-scale convolutional neural network |
| CN112184697B (en) * | 2020-10-15 | 2022-10-04 | 桂林电子科技大学 | Diabetic retinopathy grading deep learning method based on drosophila optimization algorithm |
| CN112364979B (en) * | 2020-11-05 | 2022-07-12 | 哈尔滨工业大学 | GoogLeNet-based infrared image identification method |
| CN112381821A (en) * | 2020-12-08 | 2021-02-19 | 北京青燕祥云科技有限公司 | Intelligent handheld fundus camera and image analysis method |
| CN112508919A (en) * | 2020-12-11 | 2021-03-16 | 北京大恒普信医疗技术有限公司 | Image processing method and device, electronic equipment and readable storage medium |
| CN112869697A (en) * | 2021-01-20 | 2021-06-01 | 深圳硅基智能科技有限公司 | Judgment method for simultaneously identifying stage and pathological change characteristics of diabetic retinopathy |
| CN112819797B (en) * | 2021-02-06 | 2023-09-19 | 国药集团基因科技有限公司 | Method, device, system and storage medium for analyzing diabetic retinopathy |
| CN113298083A (en) * | 2021-02-25 | 2021-08-24 | 阿里巴巴集团控股有限公司 | Data processing method and device |
| CN113011362A (en) * | 2021-03-29 | 2021-06-22 | 吉林大学 | Fine-grained fundus image grading algorithm based on bilinear pooling and attention mechanism |
| CN113378984B (en) * | 2021-07-05 | 2023-05-02 | 国药(武汉)医学实验室有限公司 | Medical image classification method, system, terminal and storage medium |
| CN113592780A (en) * | 2021-07-06 | 2021-11-02 | 南方科技大学 | Fundus image classification method, device, equipment and storage medium |
| CN113537395B (en) * | 2021-08-09 | 2022-07-08 | 同济大学 | Diabetic retinopathy image identification method based on fundus images |
| CN114898172B (en) * | 2022-04-08 | 2024-04-02 | 辽宁师范大学 | Multi-feature DAG network-based diabetic retinopathy classification modeling method |
| CN115587979B (en) * | 2022-10-10 | 2023-08-15 | 山东财经大学 | Three-stage attention network-based diabetic retinopathy grading method |
| CN116503385B (en) * | 2023-06-25 | 2023-09-01 | 吉林大学 | Sugar mesh bottom image grading method and equipment based on virtual global agent |
Citations (4)
| 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 |
| CN107239446A (en) * | 2017-05-27 | 2017-10-10 | 中国矿业大学 | A kind of intelligence relationship extracting method based on neutral net Yu notice mechanism |
| CN107291945A (en) * | 2017-07-12 | 2017-10-24 | 上海交通大学 | The high-precision image of clothing search method and system of view-based access control model attention model |
| CN107358606A (en) * | 2017-05-04 | 2017-11-17 | 深圳硅基智能科技有限公司 | For identifying the artificial neural network and system of diabetic retinopathy |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9830529B2 (en) * | 2016-04-26 | 2017-11-28 | Xerox Corporation | End-to-end saliency mapping via probability distribution prediction |
-
2017
- 2017-12-31 CN CN201711497342.9A patent/CN108021916B/en active Active
Patent Citations (4)
| 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 |
| CN107358606A (en) * | 2017-05-04 | 2017-11-17 | 深圳硅基智能科技有限公司 | For identifying the artificial neural network and system of diabetic retinopathy |
| CN107239446A (en) * | 2017-05-27 | 2017-10-10 | 中国矿业大学 | A kind of intelligence relationship extracting method based on neutral net Yu notice mechanism |
| CN107291945A (en) * | 2017-07-12 | 2017-10-24 | 上海交通大学 | The high-precision image of clothing search method and system of view-based access control model attention model |
Non-Patent Citations (1)
| Title |
|---|
| "Automated Identification of Diabetic Retinopathy Using Deep Learning";R Gargeya等;《Ophthalmology》;20170327;第124卷(第7期);第962-969页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108021916A (en) | 2018-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108021916B (en) | Deep learning diabetic retinopathy sorting technique based on attention mechanism | |
| CN109376636B (en) | Capsule network-based eye fundus retina image classification method | |
| CN109753978B (en) | Image classification method, device and computer readable storage medium | |
| CN112132817B (en) | Retina blood vessel segmentation method for fundus image based on mixed attention mechanism | |
| CN111815574B (en) | Fundus retina blood vessel image segmentation method based on rough set neural network | |
| CN109345538A (en) | A kind of Segmentation Method of Retinal Blood Vessels based on convolutional neural networks | |
| CN110276356A (en) | Eye fundus image aneurysms recognition methods based on R-CNN | |
| CN106530295A (en) | Fundus image classification method and device of retinopathy | |
| CN106530283A (en) | SVM (support vector machine)-based medical image blood vessel recognition method | |
| Peng et al. | Automatic staging for retinopathy of prematurity with deep feature fusion and ordinal classification strategy | |
| CN110084803A (en) | Eye fundus image method for evaluating quality based on human visual system | |
| WO2022088665A1 (en) | Lesion segmentation method and apparatus, and storage medium | |
| Boral et al. | Classification of diabetic retinopathy based on hybrid neural network | |
| CN114287878A (en) | Diabetic retinopathy focus image identification method based on attention model | |
| CN113012163A (en) | Retina blood vessel segmentation method, equipment and storage medium based on multi-scale attention network | |
| CN113240655A (en) | Method, storage medium and device for automatically detecting type of fundus image | |
| Lyu et al. | Deep tessellated retinal image detection using Convolutional Neural Networks | |
| Miao et al. | Classification of Diabetic Retinopathy Based on Multiscale Hybrid Attention Mechanism and Residual Algorithm | |
| Qin et al. | A review of retinal vessel segmentation for fundus image analysis | |
| Padalia et al. | A CNN-LSTM combination network for cataract detection using eye fundus images | |
| Ali et al. | Cataract disease detection used deep convolution neural network | |
| CN113011340B (en) | Cardiovascular operation index risk classification method and system based on retina image | |
| Kumari et al. | Deep learning based detection of diabetic retinopathy using retinal fundus images | |
| Xiao et al. | SE-MIDNet Based on Deep Learning for Diabetic Retinopathy Classification | |
| Khan et al. | A Computer-Aided Diagnostic System to Identify Diabetic Retinopathy, Utilizing a Modified Compact Convolutional Transformer and Low-Resolution Images to Reduce Computation Time. Biomedicines. 2023. No. 11. Art. 1566 |
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 |