CN111753820A - Color fundus image cup segmentation method based on deep learning - Google Patents
Color fundus image cup segmentation method based on deep learning Download PDFInfo
- Publication number
- CN111753820A CN111753820A CN201910236304.0A CN201910236304A CN111753820A CN 111753820 A CN111753820 A CN 111753820A CN 201910236304 A CN201910236304 A CN 201910236304A CN 111753820 A CN111753820 A CN 111753820A
- Authority
- CN
- China
- Prior art keywords
- channel
- cup
- convolution
- segmentation
- network
- 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
- 230000011218 segmentation Effects 0.000 title claims abstract description 40
- 238000013135 deep learning Methods 0.000 title claims abstract description 15
- 238000000034 method Methods 0.000 title claims abstract description 15
- 230000004044 response Effects 0.000 claims abstract description 6
- 230000005284 excitation Effects 0.000 claims description 12
- 230000006835 compression Effects 0.000 claims description 9
- 238000007906 compression Methods 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 2
- 238000012360 testing method Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 230000002207 retinal effect Effects 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 4
- 241001275117 Seres Species 0.000 description 3
- 210000004204 blood vessel Anatomy 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 208000010412 Glaucoma Diseases 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 208000020911 optic nerve disease Diseases 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 208000002177 Cataract Diseases 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000740 bleeding effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 210000000416 exudates and transudate Anatomy 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000004410 intraocular pressure Effects 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 210000001328 optic nerve Anatomy 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Images
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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The invention relates to a color fundus image cup segmentation method based on deep learning. The method comprises the following steps: 1) inputting a fundus image; 2) the method comprises the steps that a Seg-ResNet network is utilized to firstly divide a video disc, a divided video disc area is used as an interested area of the video cup division, then the Seg-ResNet network is utilized to divide the video cup of the video disc area, the network is based on a residual error basic structure, channel weighting is carried out by considering the relation among characteristic channels, the dependency relation among the channels is modeled, the characteristic response value of each channel is adjusted in a self-adaptive mode, and the characteristics of a plurality of layers are fused, so that the position information of pixel points is positioned while the image semantic information is captured; 3) and outputting the cup segmentation result by using a Seg-ResNet network. The test of the optic cup segmentation is carried out on the public data sets Glaucomarepo and Drishti-GS, and the result shows that the segmentation accuracy and the algorithm robustness are improved by the test result.
Description
Technical Field
The invention relates to a color fundus image cup segmentation method based on deep learning, which relates to image processing, deep learning and a residual error network and can segment a color fundus image cup.
Background
Glaucoma is an optic nerve disease in which optic nerve fibers are damaged due to an increase in intraocular pressure, has irreversible activity, and has a very high blinding rate, second only to cataracts. Clinically, glaucoma can be preliminarily diagnosed by measuring parameters such as cup-disc ratio and the like, and the detection of the optic cup is helpful for establishing a retinal coordinate system, so that retinal abnormalities such as drusen, exudates, bleeding points and the like and the positions of the retinal abnormalities can be further judged according to the retinal coordinate system. The automatic detection technology of the color fundus image vision cup provides a stable, accurate and efficient solution for the intelligent diagnosis of ophthalmic diseases. Therefore, in recent years, importance has been placed on the accuracy of cup segmentation of fundus images in the diagnostic study of retinal optic nerve diseases. Although the development of medical image processing technology makes the fundus image processing technology a great breakthrough and some research results are obtained in the field of fundus disease diagnosis, the accuracy and stability of the traditional medical image processing algorithm cannot be guaranteed under the conditions of complex problems and more nonlinear conditions. Factors such as small cup area, low contrast with optic discs, unobvious edge, serious blood vessel occlusion and the like in the fundus image lead the requirement of a medical image auxiliary diagnosis system to be difficult to meet by using the traditional algorithm to segment the cup in the fundus image, and a great space is still promoted.
The feature extraction based on deep learning is carried out on a large number of data sets from different shooting environments, the abstract features of the images are learned in a self-adaptive mode, the segmentation performance of different data sets is stable, and the problems that the traditional method is complex in operation and low in robustness are solved. The deep learning segmentation framework solves the problem of semantic level image segmentation, so that each pixel in an image can be subjected to target and background binary classification, and the prediction result of each pixel is predicted end to end. Compared with the traditional technology, the deep learning-based optic cup segmentation method has made a great breakthrough, but due to the particularity of the medical image, the requirement of a large-scale data set required by deep network training is difficult to meet, so that the current deep learning-based optic cup segmentation precision is not high.
Disclosure of Invention
The invention aims to improve the segmentation precision of the optic cup in a color fundus image, provides an optic cup segmentation method based on a deep learning technology by combining a transfer learning strategy aiming at the existing small-scale data set, and adopts the following technical scheme for the invention:
1. inputting a fundus image, and extracting a region-of-interest video area as a cup segmentation area;
2. performing cup segmentation on the video area;
3. optimizing a residual error structure;
4. adjusting the response value of each characteristic channel by adopting a channel weighting structure;
5. fusing the high-level features with the bottom-level features;
6. and outputting the cup segmentation result by using a Seg-ResNet network.
In the step 1: because the optic cup area is small and is close to the brightness characteristic of the optic disc, in order to avoid the interference of the optic disc on the optic cup segmentation, the optic disc needs to be segmented firstly before the optic cup is segmented, and the optic disc area is used as the interesting area of the optic cup segmentation.
In the step 2: and performing cup segmentation by taking the segmented optic disc area as an interesting area.
In the step 3: optimizing residual block computation, replacing 3 × 3 two-layer convolution with 1 × 1+3 × 3+1 × 1 three-layer convolution, firstly reducing dimensionality under one 1 × 1 convolution filtering to reduce computation, optimizing the middle 3 × 3 convolution layer in the structure, and finally restoring dimensionality through another 1 × 1 convolution filtering.
In the step 4: in the compression stage, the two-dimensional feature map output by each layer is compressed to one dimension by using global average pooling, and the one-dimensional features obtained by compression can reflect global receptive field information of the input feature channels; in the excitation stage, learnable weights are generated for the compressed characteristic channels, and the nonlinear modeling can be explicitly carried out on the correlation among the characteristic channels through excitation; the obtained output weight represents the importance degree of each characteristic channel after compression and excitation in the steps, a group of characteristics are directly output after network convolution in the weighting stage, and then the weight obtained by excitation is weighted to the directly output characteristics channel by channel through multiplication, so that the original characteristic channel is reweighed.
In the step 5: using convolution of 1 × 1 to reduce the dimension of the high-dimension feature map in the encoding stage to the same dimension as the feature map in the decoding stage for fusion operation, using convolution kernel of 1 × 1 to reduce the dimension of the feature map, and then performing feature fusion on the feature map with the same size in the bottom layer in the same manner as described above.
In the step 6: and detecting the data sets Glaucomarepo and Drishti-GS by using a Seg-ResNet network, and outputting a segmentation result.
Compared with the prior art, the invention has the beneficial effects that:
1. avoiding gradient explosion or gradient disappearance. The identity mapping superimposes the input onto the convolution output, skipping the concatenation of layers.
2. And (6) optimizing and calculating. The three-layer convolution of 1 × 1+3 × 3+1 × 1 is used instead of the two-layer convolution of 3 × 3.
3. Simple structure and easy realization. And explicitly carrying out nonlinear modeling on the correlation among the channels by considering the relation among the channels of the feature diagram after convolution.
4. The segmentation precision and robustness of the optic cup are improved.
Drawings
FIG. 1 is a diagram of a deep learning network structure Seg-ResNet;
FIG. 2 is a diagram of an optimized residual structure;
FIG. 3 is a schematic diagram of an SE module;
FIG. 4 is a high-level feature and low-level feature fusion graph;
FIGS. 5(a1) - (a3) are graphs of segmentation results in the Glaucomara database severely occluded by blood vessels;
FIGS. 5(b1) - (b3) are graphs of segmentation results with low cup-to-disk contrast in the Glaucomarepo database;
FIGS. 5(c1) - (c3) are graphs of segmentation results of Glaucomara database with different sizes of view cup areas;
FIGS. 5(d1) - (d3) are the segmentation results of the Drishti-GS database with severe occlusion by blood vessels;
FIGS. 5(e1) - (e3) are graphs showing the segmentation results of low cup-to-disk contrast in the Drishti-GS database;
FIGS. 5(d1) - (d3) are diagrams of segmentation results in the Drishti-GS database with different sizes of view cup areas;
Detailed Description
The deep learning network structure of the invention is shown in fig. 1, and the Seg-ResNet comprises 10 SERes modules, 5 maximum pooling layers, 4 upsampling layers and a loss function layer. The convolution kernel in the network has two sizes, 1 × 1 and 3 × 3, respectively, with a step size of 1, and each layer of convolution is followed by activation using the ReLU function to increase the non-linear capability of the network. The convolution size of the maximum pooling layer is 2 x 2 with a step size of 2. SERes in the network structure are modules combining a residual error structure and a channel weighting structure, each SERes module comprises three convolution layers, and a BatchNorm layer is used in convolution to carry out batch normalization processing on data. Firstly, using ResNet as a basic network, adopting a channel weighting structure to adjust the response value of each characteristic channel, and fusing the high-level characteristics and the bottom-level characteristics to form a new network Seg-ResNet. The following describes a specific implementation process of the technical solution of the present invention with reference to the accompanying drawings.
1. Inputting the fundus image, extracting the region of interest
The video area is used as a region for cup segmentation, and the video area of the region of interest is extracted by using data in public data sets Glaucomarepo and Drishti-GS as input fundus images.
2. And performing cup segmentation on the video disc area, and performing cup segmentation on the obtained video disc area as an interested area.
3. Optimized residual structure
The residual structure model may utilize identity mapping to degenerate the deep network into a shallow network. The optimized residual structure is shown in fig. 2.
4. Adjusting response value of each characteristic channel by adopting channel weighting structure
The channel weighting is to look for the dependency relationship of each channel in consideration of the dimension of the feature map, model the importance degree of different feature channels in a learning mode, selectively enhance the main features and inhibit the unimportant features. The output of the conventional neural network ignores the dependency relationship among channels, and the Seg-ResNet is designed based on a channel weighting unit module, so that useful features in a feature diagram obtained by the network convolution are more fully utilized, and the useless features are suppressed. In the compression stage, the two-dimensional feature map output by each layer is compressed to one dimension by using global average pooling, and the one-dimensional features obtained by compression can reflect global receptive field information of the input feature channels; the excitation phase generates learnable weights for the compressed feature channels, similar to the mechanism of gates in a recurrent neural network. The correlation among the characteristic channels can be explicitly modeled in a non-linear mode through excitation; the weighting stage passes through a compression stage and an excitation stage, the obtained output weight represents the importance degree of each characteristic channel, after network convolution, a group of characteristics are directly output, then the weight obtained by excitation is weighted to the directly output characteristics channel by channel through multiplication, and the re-weighting of the original characteristic channel is completed. A schematic diagram of the SE module is shown in fig. 3.
5. Fusing high-level features with underlying features
(1) And restoring the characteristic graph size in the decoding stage. In the decoding stage, the feature map compressed in the encoding stage is restored through continuous up-sampling so as to enable the feature map to be consistent with the size of the bottom layer feature map to be fused;
(2) and reducing the dimension of the feature map in the encoding stage. The feature graph dimension of the encoding stage is higher than that of the decoding stage, and the feature graph output by the bottom layer of the 1 multiplied by 1 convolution kernel dimension reduction encoding stage is used for being conveniently fused with the feature graph recovered by the high layer;
(3) high-level and low-level features are fused. And fusing through the Concat layer after the sizes and dimensions of the characteristic graphs of the upper layer and the bottom layer are consistent. The high-level feature and low-level feature fusion map is shown in fig. 4.
6. Outputting cup segmentation result by using Seg-ResNet network
The parameters of the view cup segmentation network are set as follows: and optimizing training by adopting a random gradient descent algorithm with momentum, wherein the momentum value is 0.95, the learning rate is set to be 0.001, and the epoch value is 12.
The test image is input into the network and the final segmentation result of the optic cup is obtained as shown in fig. 5.
The image recognition method based on deep learning is used for cup segmentation, and the deep learning algorithm is adopted to train and analyze the color fundus image data, so that cup segmentation as accurate as possible can be realized. The technology is applied to medicine, and the intelligent diagnosis of ophthalmic diseases is greatly facilitated.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. All within the spirit and principles of the present invention. Any modifications, equivalents and the like which are made are intended to be included within the scope of the present invention.
Claims (4)
1. The color fundus image cup segmentation method based on deep learning comprises the following steps:
(1) inputting a fundus image, and extracting a region-of-interest video area as a cup segmentation area;
(2) performing cup segmentation on the video area;
(3) optimizing a residual error structure;
(4) adjusting the response value of each characteristic channel by adopting a channel weighting structure;
(5) fusing the high-level features with the bottom-level features;
(6) and outputting the cup segmentation result by using a Seg-ResNet network.
2. The method of claim 1, wherein in step (3), ResNet based network is used to optimize residual block computation, and 1 x 1+3 x 3+1 x 1 tri-layer convolution is used to replace 3 x 3 bi-layer convolution, with dimensionality reduction computation being first performed under one 1 x 1 convolution filter, optimizing the middle 3 x 3 convolution layer in the structure, and dimensionality reduction being finally performed by another 1 x 1 convolution filter.
3. The method according to claim 1, wherein in step (4), a channel weighting structure is adopted to adjust the response value of each feature channel, the compression stage compresses the two-dimensional feature map output by each layer to one dimension by using global average pooling, and the one-dimensional features obtained by compression reflect global receptive field information of the input feature channels; in the excitation stage, learnable weights are generated for the compressed characteristic channels, and the nonlinear modeling is performed on the correlation among the characteristic channels through excitation explicit mode; the obtained output weight represents the importance degree of each characteristic channel after compression and excitation in the steps, a group of characteristics are directly output after network convolution in the weighting stage, and then the weight obtained by excitation is weighted to the directly output characteristics channel by channel through multiplication, so that the original characteristic channel is reweighed.
4. The method of claim 1, wherein in step (5), the high-level features are fused with the low-level features, the high-dimensional feature map in the encoding stage is reduced in dimension by using a1 x 1 convolution, the dimension is reduced to the same dimension as the feature map in the decoding stage for the fusion operation, the fusion is followed by reducing the dimension of the feature map by using a1 x 1 convolution kernel, and the feature fusion is then performed on the low-level feature map with the same dimension in the same manner as described above.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910236304.0A CN111753820A (en) | 2019-03-27 | 2019-03-27 | Color fundus image cup segmentation method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910236304.0A CN111753820A (en) | 2019-03-27 | 2019-03-27 | Color fundus image cup segmentation method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111753820A true CN111753820A (en) | 2020-10-09 |
Family
ID=72671883
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910236304.0A Pending CN111753820A (en) | 2019-03-27 | 2019-03-27 | Color fundus image cup segmentation method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111753820A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111986202A (en) * | 2020-10-26 | 2020-11-24 | 平安科技（深圳）有限公司 | Glaucoma auxiliary diagnosis device, method and storage medium |
| CN112686297A (en) * | 2020-12-29 | 2021-04-20 | 中国人民解放军海军航空大学 | Radar target motion state classification method and system |
| CN115100187A (en) * | 2022-07-27 | 2022-09-23 | 浙江大学 | Glaucoma image detection method based on federal learning |
| CN115546217A (en) * | 2022-12-02 | 2022-12-30 | 中南大学 | Multi-level fusion skin disease diagnosis system based on multi-mode image data |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108510473A (en) * | 2018-03-09 | 2018-09-07 | 天津工业大学 | The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth |
| CN109389667A (en) * | 2018-08-23 | 2019-02-26 | 北京大学 | A kind of efficient global illumination method for drafting based on deep learning |
| CN109410129A (en) * | 2018-09-28 | 2019-03-01 | 大连理工大学 | A kind of method of low light image scene understanding |
| CN109447962A (en) * | 2018-10-22 | 2019-03-08 | 天津工业大学 | A kind of eye fundus image hard exudate lesion detection method based on convolutional neural networks |
-
2019
- 2019-03-27 CN CN201910236304.0A patent/CN111753820A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108510473A (en) * | 2018-03-09 | 2018-09-07 | 天津工业大学 | The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth |
| CN109389667A (en) * | 2018-08-23 | 2019-02-26 | 北京大学 | A kind of efficient global illumination method for drafting based on deep learning |
| CN109410129A (en) * | 2018-09-28 | 2019-03-01 | 大连理工大学 | A kind of method of low light image scene understanding |
| CN109447962A (en) * | 2018-10-22 | 2019-03-08 | 天津工业大学 | A kind of eye fundus image hard exudate lesion detection method based on convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| ZHITAO XIAO ET AL.: "Optic cup segmentation method by a modified VGG-16 network", 《JOURNAL OF MEDICAL IMAGING AND HEALTH INFORMATICS》, pages 97 - 101 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111986202A (en) * | 2020-10-26 | 2020-11-24 | 平安科技（深圳）有限公司 | Glaucoma auxiliary diagnosis device, method and storage medium |
| CN112686297A (en) * | 2020-12-29 | 2021-04-20 | 中国人民解放军海军航空大学 | Radar target motion state classification method and system |
| CN112686297B (en) * | 2020-12-29 | 2023-04-14 | 中国人民解放军海军航空大学 | Radar target motion state classification method and system |
| CN115100187A (en) * | 2022-07-27 | 2022-09-23 | 浙江大学 | Glaucoma image detection method based on federal learning |
| CN115100187B (en) * | 2022-07-27 | 2024-06-11 | 浙江大学 | Glaucoma image detection method based on federal learning |
| CN115546217A (en) * | 2022-12-02 | 2022-12-30 | 中南大学 | Multi-level fusion skin disease diagnosis system based on multi-mode image data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111145170B (en) | Medical image segmentation method based on deep learning | |
| CN111753820A (en) | Color fundus image cup segmentation method based on deep learning | |
| WO2022036777A1 (en) | Method and device for intelligent estimation of human body movement posture based on convolutional neural network | |
| CN111950649B (en) | Attention mechanism and capsule network-based low-illumination image classification method | |
| CN106920227B (en) | The Segmentation Method of Retinal Blood Vessels combined based on deep learning with conventional method | |
| CN112508864B (en) | Retinal vessel image segmentation method based on improved UNet + | |
| CN110503063B (en) | Falling detection method based on hourglass convolution automatic coding neural network | |
| CN114998210B (en) | Retinopathy of prematurity detecting system based on deep learning target detection | |
| CN112132817A (en) | Retina blood vessel segmentation method for fundus image based on mixed attention mechanism | |
| CN113888412B (en) | Image super-resolution reconstruction method for diabetic retinopathy classification | |
| Harshitha et al. | Predicting the stages of diabetic retinopathy using deep learning | |
| Xu et al. | Dual-channel asymmetric convolutional neural network for an efficient retinal blood vessel segmentation in eye fundus images | |
| CN114494195A (en) | Small sample attention mechanism parallel twinning method for fundus image classification | |
| CN113012163A (en) | Retina blood vessel segmentation method, equipment and storage medium based on multi-scale attention network | |
| CN112580545A (en) | Crowd counting method and system based on multi-scale self-adaptive context network | |
| CN110610480B (en) | MCASPP neural network eyeground image optic cup optic disc segmentation model based on Attention mechanism | |
| Radha et al. | Modified Depthwise Parallel Attention UNet for Retinal Vessel Segmentation | |
| Pradhan et al. | Diabetic retinopathy detection on retinal fundus images using convolutional neural network | |
| Haider et al. | Modified Anam-Net Based Lightweight Deep Learning Model for Retinal Vessel Segmentation. | |
| CN117409358A (en) | BiFPN-fused light flame detection method | |
| CN116884036A (en) | Live pig posture detection method, device, equipment and medium based on YOLOv5DA | |
| Khan et al. | Screening fundus images to extract multiple ocular features: A unified modeling approach | |
| CN110992320A (en) | Medical image segmentation network based on double interleaving | |
| Zijian et al. | AFFD-Net: A Dual-Decoder Network Based on Attention-Enhancing and Feature Fusion for Retinal Vessel Segmentation | |
| Joans | Identification and classification of eye disease using deep learning |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20201009 |