CN108921227B - Glaucoma medical image classification method based on capsule theory - Google Patents

Glaucoma medical image classification method based on capsule theory Download PDF

Info

Publication number
CN108921227B
CN108921227B CN201810759934.1A CN201810759934A CN108921227B CN 108921227 B CN108921227 B CN 108921227B CN 201810759934 A CN201810759934 A CN 201810759934A CN 108921227 B CN108921227 B CN 108921227B
Authority
CN
China
Prior art keywords
layer
image
glaucoma
capsule
capsules
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
Application number
CN201810759934.1A
Other languages
Chinese (zh)
Other versions
CN108921227A (en
Inventor
刘少鹏
贾西平
关立南
高维奇
洪佳明
李耿鑫
张倩
林智勇
崔怀林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Polytechnic Normal University
Original Assignee
Guangdong Polytechnic Normal University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangdong Polytechnic Normal University filed Critical Guangdong Polytechnic Normal University
Priority to CN201810759934.1A priority Critical patent/CN108921227B/en
Publication of CN108921227A publication Critical patent/CN108921227A/en
Application granted granted Critical
Publication of CN108921227B publication Critical patent/CN108921227B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Abstract

The invention discloses a glaucoma medical image classification method based on capsule theory, which comprises the steps of image preprocessing operation, construction of a convolutional neural network based on capsule and construction of an image reconstruction full-connection network; the image preprocessing operation firstly cuts out an interested area by utilizing the existing glaucoma medical image optic disc detection model to reduce interference information, then fuses optic disc and optic cup semantic segmentation graph information, and finally adopts a self-adaptive histogram for limiting the contrast to enhance, thereby improving the integral or local contrast of the image. The convolutional neural network based on capsule not only can automatically learn the medical image characteristics of glaucoma, but also can find the position and direction information among the characteristics, thereby identifying the glaucoma more accurately. The image reconstruction full-connection network uses the category capsules to recover the original medical image of glaucoma, and aims to improve the generalization capability of the convolution neural network based on the capsules.

Description

Glaucoma medical image classification method based on capsule theory
Technical Field
The invention relates to the field of glaucoma medical image analysis, in particular to a glaucoma medical image classification method based on capsule theory.
Background
Glaucoma is a group of diseases characterized by common atrophy of the optic nerve, visual field defects, and vision loss. If the treatment is not performed in time, blindness can be caused. Glaucoma has become one of three major cases of blindness, with a total incidence of 1% and 2% after age 45. The early screening of glaucoma has great significance, and is beneficial to timely treatment of patients, thereby protecting eyesight and even healing.
Because the number of potential glaucoma patients is large, limited medical resources and clinical experience cannot meet the existing requirements, and the development of early screening work by ophthalmologists is very difficult. The computer-assisted glaucoma early screening can effectively relieve the working pressure of ophthalmologists and provide better medical service for patients.
Common computer techniques for glaucoma medical image classification include traditional computer vision techniques and deep learning. The traditional computer vision technology extracts structural information such as image texture features and gray scale features, and then the traditional machine learning method including naive Bayes, K nearest neighbor, support vector machine, random forest and the like is used for finally completing the glaucoma classification task. Because the feature selection depends on manual experience, abstract semantic feature information cannot be automatically found, and the model is difficult to generalize. Most model training samples are small in scale, and the practical application value is not high.
The deep learning is an important branch of artificial intelligence, can automatically learn sample characteristics without manual participation, and is a complete end-to-end model. Deep learning has made a series of breakthrough progress in image classification and other tasks and is widely applied. The classification and identification of medical images of glaucoma in combination with deep learning becomes a research hotspot, and particularly, several research results have been obtained by using Convolutional Neural Networks (CNN). The method comprises the steps of firstly selecting an interested region of the glaucoma medical image, then carrying out image preprocessing to be used as CNN input, automatically learning high-dimensional semantic features of the image after a series of convolution, pooling and activation operations, and finally completing classification of the glaucoma medical image by subsequent full-connection network fusion features. Although CNN is able to automatically learn features, it cannot identify spatial relationships between features. In order to better model the hierarchical relationship (including position, direction and the like) of internal knowledge representation in the network, Hinton proposes a capsule theory, designs a Capsule model on the basis of the capsule theory, and obtains an image classification effect superior to CNN on a public experimental data set. At present, there is no research related to assisting glaucoma screening by using capsule theory for a while. In addition, the existing deep learning method is directly adopted to process the medical image of glaucoma, and the specific knowledge in the field of glaucoma, such as information of a optic disc and a optic cup segmentation map of an eye fundus map, is ignored, so that the classification accuracy of glaucoma is low.
Disclosure of Invention
The invention aims to solve one or more of the defects and provides a glaucoma medical image classification method based on capsule theory.
In order to realize the purpose, the technical scheme is as follows:
a glaucoma medical image classification method based on capsule theory comprises the following steps:
s1: carrying out preprocessing operation on the image;
s2: constructing a convolutional neural network based on capsule, inputting the preprocessed image, and outputting a glaucoma recognition result;
s3: and constructing an image reconstruction full-connection network, and recovering the original image by using the category capsule.
Preferably, step S1 includes the steps of:
s1.1: cutting out an interested area by using the existing glaucoma medical image optic disc detection model to reduce interference information;
s1.2: fusing the information of the video disc and the video cup semantic segmentation map;
s1.3: and (3) adopting self-adaptive histogram enhancement for limiting the contrast so as to improve the overall or local contrast of the image.
Preferably, the convolutional neural network based on capsule described in step S2 includes a first layer of convolutional layer ReLU Conv1, a second layer of Primary Caps and a third layer of Glaucoma Caps;
the first layer is a convolutional layer ReLU Conv1, which is a common convolutional layer, the size of an input image is 4 × 28 × 28, that is, 3 RGB channel information of an original image and single-channel labeling information of a video disc and video cup segmentation graph are combined, the layer has 256 convolution kernels of 9 × 9, the step length is 1, an activation function is ReLU, local feature extraction is realized, the original image information and the video disc and video cup labeling information are fused, and the input image is used as the input of the next layer;
a second Primary Caps layer which is a capsule layer; the input image size is 256 × 20 × 20, the vector dimension of the capsule vector of the layer is set to be 8, 32 convolution kernels of 9 × 9 are shared, the step size is 2, the activation function is ReLU, the output is 32 channels, and the data size of each channel is 6 × 6;
the third layer is a Glaucoma Caps layer which is a capsule layer; the input data is 256 × 6 × 6 × 8, the vector dimension of the capsule in the layer is set to be 16, and 2 classes of capsules are output and respectively correspond to the glaucoma identification results, namely normal glaucoma and glaucoma.
Preferably, the convolutional neural network based on capsule adopts a back propagation technology, and minimizes the loss function L through parameter updating;
where the loss function L of the entire network:
L＝αLc+βLr
Lcrepresenting the probability of existence of the capsule entity by using the length of the vector as a marginal loss function; l isrTo reconstruct losses; α and β represent the weight of the margin loss and reconstruction loss, respectively;
wherein
Lc＝Tc·[max(0,m+-||Vc||)]2+λ·(1-Tc)·[max(0,||Vc||-m-)]2；
VcIs the capsule vector of glaucoma image class c, | VcI denotes the length of the vector, TcIf and only if a cyan-eye image class c exists, m+＝0.9,m-0.1, λ represents the weight attenuation coefficient of the existing class;
Lr＝||x-x'||2；
x is an original image (including labeling information of a video disc and a video cup segmentation graph), and x' is a reconstructed image; | | x-x' | purple light2The difference between the two images is represented by the following specific calculation method: firstly, the difference of the gray values of the pixel points at the same positions of the two images is calculated, then the square of the difference value is obtained, and finally the result is obtained by summation.
Preferably, the image reconstruction full-connection network described in step S3 uses class capsules to restore the original image; wherein the first layer is Glaucoma Caps, the registration soft retrieval target represents selected capsules, and the mask marks unselected capsules; the second layer is a full-connection layer, 2048 neurons are totally formed, and the activation function is ReLU; the third layer is a full connection layer, 4096 neurons are increased, and the activation function is ReLU; the final output layer is a fully connected layer with 3136 neurons, corresponding to a 3 × 28 × 28 original image, and a 1 × 28 × 28 optic disc and cup segmented image.
Preferably, the image reconstruction fully-connected network passes through a square loss L between the network output and the original image inputrAnd (6) calculating. Updating network parameters using back propagation techniques such that LrAnd (4) minimizing.
Compared with the prior art, the invention has the beneficial effects that:
1) the method can effectively integrate the special knowledge in the field of glaucoma and the capsule theory, and has better accuracy than the prior CNN method in the medical image classification task of glaucoma;
2) the network has good training stability and rapid convergence.
Drawings
FIG. 1 is a framework of a glaucoma medical image classification network based on capsule theory;
FIG. 2 is a convolutional neural network based on capsules;
FIG. 3 is an image reconstruction fully connected network;
FIG. 4 is a training process for a capsule based convolutional neural network;
fig. 5 is a glaucoma classification procedure.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
the invention is further illustrated below with reference to the figures and examples.
Example 1
A glaucoma medical image classification method based on capsule theory comprises the following steps:
s1: carrying out preprocessing operation on the image;
s2: constructing a convolutional neural network based on capsule, inputting the preprocessed image, and outputting a glaucoma recognition result;
s3: and constructing an image reconstruction full-connection network, and recovering the original image by using the category capsule.
In this embodiment, step S1 includes the following steps:
s1.1: cutting out an interested area by using the existing glaucoma medical image optic disc detection model to reduce interference information;
s1.2: fusing the information of the video disc and the video cup semantic segmentation map;
s1.3: and (3) adopting self-adaptive histogram enhancement for limiting the contrast so as to improve the overall or local contrast of the image.
In this embodiment, the convolutional neural network based on capsule described in step S2 includes a first convolutional layer ReLU Conv1, a second convolutional Caps layer, and a third Glaucoma Caps layer, please refer to fig. 2;
the first layer is a convolutional layer ReLU Conv1, which is a common convolutional layer, the size of an input image is 4 × 28 × 28, that is, 3 RGB channel information of an original image and single-channel labeling information of a video disc and video cup segmentation graph are combined, the layer has 256 convolution kernels of 9 × 9, the step length is 1, an activation function is ReLU, local feature extraction is realized, the original image information and the video disc and video cup labeling information are fused, and the input image is used as the input of the next layer;
a second Primary Caps layer which is a capsule layer; the input image size is 256 × 20 × 20, the vector dimension of the capsule vector of the layer is set to be 8, 32 convolution kernels of 9 × 9 are shared, the step size is 2, the activation function is ReLU, the output is 32 channels, and the data size of each channel is 6 × 6;
the third layer is a Glaucoma Caps layer which is a capsule layer; the input data is 256 × 6 × 6 × 8, the vector dimension of the capsule in the layer is set to be 16, and 2 classes of capsules are output and respectively correspond to the glaucoma identification results, namely normal glaucoma and glaucoma.
In this embodiment, the convolutional neural network based on capsule adopts a back propagation technique, and minimizes a loss function L by updating parameters;
where the loss function L of the entire network:
L＝αLc+βLr
Lcrepresenting the probability of existence of the capsule entity by using the length of the vector as a marginal loss function; l isrTo reconstruct losses; α and β represent the weight of the margin loss and reconstruction loss, respectively;
wherein
Lc＝Tc·[max(0,m+-||Vc||)]2+λ·(1-Tc)·[max(0,||Vc||-m-)]2；
VcIs the capsule vector of glaucoma image class c, | VcI denotes the length of the vector, TcIf and only if a cyan-eye image class c exists, m+＝0.9,m-0.1, λ represents the weight attenuation coefficient of the existing class;
Lr＝||x-x'||2；
x is an original image (including labeling information of a video disc and a video cup segmentation graph), and x' is a reconstructed image; | | x-x' | purple light2The difference between the two images is represented by the following specific calculation method: firstly, the difference of the gray values of the pixel points at the same positions of the two images is calculated, then the square of the difference value is obtained, and finally the result is obtained by summation.
The training process is shown in fig. 4. The horizontal axis is the training round, the vertical axis is the loss function value, the lower curve is the training loss, and the upper curve is the testing loss. As can be seen from the graph, the training loss was always in a downward trend in 300 experiments, while the test loss was reduced in about 50 experiments and then substantially stabilized. Therefore, the network has good training stability and rapid convergence.
In this embodiment, referring to fig. 3, the image reconstruction full-connection network described in step S3 uses class capsules to restore the original image; wherein the first layer is Glaucoma Caps, the registration of the registration target indicates selected capsules, and Masked marks unselected capsules; the second layer is a full-connection layer, 2048 neurons are totally formed, and the activation function is ReLU; the third layer is a full connection layer, 4096 neurons are increased, and the activation function is ReLU; the final output layer is a fully connected layer with 3136 neurons, corresponding to a 3 × 28 × 28 original image, and a 1 × 28 × 28 optic disc and cup segmented image.
In this embodiment, the image reconstruction fully-connected network passes through a square loss L between the network output and the original image inputrAnd (6) calculating. Updating network parameters using back propagation techniques such that LrAnd (4) minimizing.
Example 2
Giving an eye fundus image and information of a video disc and a video cup segmentation icon of the eye fundus image, processing the eye fundus image and the information by adopting an image preprocessing technology, using the image preprocessing technology as input of a trained semantic segmentation network, outputting classification probability through operation, and judging whether an original image is glaucoma or not, wherein the specific flow is shown in fig. 5.
S1: reading the labeling information of the fundus picture and the optic disc and cup segmentation picture thereof;
s2: fusing the fundus image and the segmentation image thereof, and overlapping the original 3-channel RGB fundus image and the single-channel segmentation image into a new 4-channel image;
s3: cutting the optic disc area of the new 4-channel image to obtain an interested area with the size of 4 multiplied by 28;
s4: processing the region of interest by using CLAHE to enhance the image or local contrast;
s5: taking the preprocessed image as the input of a trained convolutional neural network based on capsules;
s6: the network outputs the classification probability value so as to judge whether the original image is glaucoma.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

Claims (4)

1. A glaucoma medical image classification method based on capsule theory is characterized by comprising the following steps:
s1: carrying out preprocessing operation on the image;
step S1 includes the following steps:
s1.1: cutting out an interested area by using the existing glaucoma medical image optic disc detection model to reduce interference information;
s1.2: fusing the information of the video disc and the video cup semantic segmentation map;
s1.3: the contrast is limited by adopting self-adaptive histogram enhancement, so that the overall or local contrast of the image is improved;
s2: constructing a convolutional neural network based on capsule, inputting the preprocessed image, and outputting a glaucoma recognition result; the convolutional neural network based on capsules of step S2 includes a first layer of convolutional layer ReLU Conv1, a second layer of Primary Caps and a third layer of Glaucoma Caps;
the first layer is a convolutional layer ReLU Conv1, which is a common convolutional layer, the size of an input image is 4 × 28 × 28, that is, 3 RGB channel information of an original image and single-channel labeling information of a video disc and video cup segmentation graph are combined, the layer has 256 convolution kernels of 9 × 9, the step length is 1, an activation function is ReLU, local feature extraction is realized, the original image information and the video disc and video cup labeling information are fused, and the input image is used as the input of the next layer;
a second Primary Caps layer which is a capsule layer; the input image size is 256 × 20 × 20, the vector dimension of the capsule vector of the layer is set to be 8, 32 convolution kernels of 9 × 9 are shared, the step size is 2, the activation function is ReLU, the output is 32 channels, and the data size of each channel is 6 × 6;
the third layer is a Glaucoma Caps layer which is a capsule layer; the input data is 256 multiplied by 6 multiplied by 8, the vector dimension of the capsule in the layer is set as 16, 2 classes of capsules are output, and the classes of capsules respectively correspond to the recognition result of the glaucoma, namely normal glaucoma and glaucoma;
s3: and constructing an image reconstruction full-connection network, and recovering the original image by using the category capsule.
2. The method for classifying medical images of glaucoma based on capsule theory as claimed in claim 1, wherein the convolutional neural network based on capsule adopts a back propagation technique to minimize the loss function L by updating parameters;
where the loss function L of the entire network:
L＝αLc+βLr
α and β represent the weight of the margin loss and reconstruction loss, respectively; l iscRepresenting the probability of existence of the capsule entity by using the length of the vector as a marginal loss function; l isrTo reconstruct losses;
wherein
Lc＝Tc·[max(0,m+-||Vc||)]2+λ·(1-Tc)·[max(0,||Vc||-m-)]2；
VcIs the capsule vector of glaucoma image class c, | VcI denotes the length of the vector, TcIf and only if a cyan-eye image class c exists, m+＝0.9,m-0.1, λ represents the weight attenuation coefficient of the existing class;
Lr＝||x-x'||2；
x is an original image (including labeling information of a video disc and a video cup segmentation graph), and x' is a reconstructed image; | | x-x' | purple light2The difference between the two images is represented by the following specific calculation method: firstly, the difference of the gray values of the pixel points at the same positions of the two images is calculated, then the square of the difference value is obtained, and finally the result is obtained by summation.
3. The method for classifying medical images for glaucoma based on capsule theory as claimed in claim 1, wherein the image reconstruction full-connection network of step S3 restores the original image using the capsule class; wherein the first layer is Glaucoma Caps, the registration of the registration target indicates selected capsules, and Masked marks unselected capsules; the second layer is a full connection layer, 2048 neurons are totally formed, and the activation function is ReLU; the third layer is a full connection layer, 4096 neurons are increased, and the activation function is ReLU; the final output layer is a fully connected layer with 3136 neurons, corresponding to a 3 × 28 × 28 original image, and a 1 × 28 × 28 optic disc and cup segmented image.
4. The method for classifying medical images for glaucoma based on capsule theory as claimed in claim 2 or 3, wherein the image reconstruction full-connection network passes through a square loss L between a network output and an original image inputrAnd (6) calculating. Updating network parameters using back propagation techniques such that LrAnd (4) minimizing.
CN201810759934.1A 2018-07-11 2018-07-11 Glaucoma medical image classification method based on capsule theory Active CN108921227B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810759934.1A CN108921227B (en) 2018-07-11 2018-07-11 Glaucoma medical image classification method based on capsule theory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810759934.1A CN108921227B (en) 2018-07-11 2018-07-11 Glaucoma medical image classification method based on capsule theory

Publications (2)

Publication Number Publication Date
CN108921227A CN108921227A (en) 2018-11-30
CN108921227B true CN108921227B (en) 2022-04-08

Family

ID=64412438

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810759934.1A Active CN108921227B (en) 2018-07-11 2018-07-11 Glaucoma medical image classification method based on capsule theory

Country Status (1)

Country Link
CN (1) CN108921227B (en)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109711411B (en) * 2018-12-10 2020-10-30 浙江大学 Image segmentation and identification method based on capsule neurons
CN110211117B (en) * 2019-05-31 2023-08-15 广东世纪晟科技有限公司 Processing system for identifying linear tubular objects in medical image and optimized segmentation method
CN110288555B (en) * 2019-07-02 2022-08-02 桂林电子科技大学 Low-illumination enhancement method based on improved capsule network
CN112329497A (en) * 2019-07-18 2021-02-05 杭州海康威视数字技术股份有限公司 Target identification method, device and equipment
CN110458852B (en) * 2019-08-13 2022-10-21 四川大学 Lung tissue segmentation method, device and equipment based on capsule network and storage medium
CN110619639A (en) * 2019-08-26 2019-12-27 苏州同调医学科技有限公司 Method for segmenting radiotherapy image by combining deep neural network and probability map model
CN110599491B (en) * 2019-09-04 2024-04-12 腾讯医疗健康(深圳)有限公司 Priori information-based eye image segmentation method, apparatus, device and medium
CN113012093B (en) * 2019-12-04 2023-12-12 深圳硅基智能科技有限公司 Training method and training system for glaucoma image feature extraction
WO2021128230A1 (en) * 2019-12-27 2021-07-01 上海昕健医疗技术有限公司 Deep learning-based medical image processing method and system, and computer device
CN111210414B (en) * 2020-01-06 2023-12-26 上海联影智能医疗科技有限公司 Medical image analysis method, computer device, and readable storage medium
CN111667490B (en) * 2020-05-07 2023-06-30 清华大学深圳国际研究生院 Fundus picture cup optic disc segmentation method
CN113052012B (en) * 2021-03-08 2021-11-19 广东技术师范大学 Eye disease image identification method and system based on improved D-S evidence
CN113705613B (en) * 2021-07-27 2024-02-02 浙江工业大学 X-ray radius distal fracture classification method based on spatial position guidance

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104955388A (en) * 2012-11-13 2015-09-30 艾欧敏达有限公司 Neurophysiological data analysis using spatiotemporal parcellation
CN106934365A (en) * 2017-03-09 2017-07-07 广东顺德中山大学卡内基梅隆大学国际联合研究院 A kind of reliable glaucoma patient self-detection method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104955388A (en) * 2012-11-13 2015-09-30 艾欧敏达有限公司 Neurophysiological data analysis using spatiotemporal parcellation
CN106934365A (en) * 2017-03-09 2017-07-07 广东顺德中山大学卡内基梅隆大学国际联合研究院 A kind of reliable glaucoma patient self-detection method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Convolutional capsule network for classification of breast cancer histology images;Tomas Iesmantas 等;《International Conference on Image Analysis and Recognition 2018》;20180430;正文第2-4小节，图1-2 *

Also Published As

Publication number Publication date
CN108921227A (en) 2018-11-30

Similar Documents

Publication Publication Date Title
CN108921227B (en) Glaucoma medical image classification method based on capsule theory
Asiri et al. Deep learning based computer-aided diagnosis systems for diabetic retinopathy: A survey
CN110992382B (en) Fundus image optic cup optic disc segmentation method and system for assisting glaucoma screening
Hacisoftaoglu et al. Deep learning frameworks for diabetic retinopathy detection with smartphone-based retinal imaging systems
Chakrabarty A deep learning method for the detection of diabetic retinopathy
EP3659067B1 (en) Method of modifying a retina fundus image for a deep learning model
Haloi Improved microaneurysm detection using deep neural networks
CN112132817B (en) Retina blood vessel segmentation method for fundus image based on mixed attention mechanism
CN109615632B (en) Fundus image optic disc and optic cup segmentation method based on semi-supervision condition generation type countermeasure network
Ajitha et al. Identification of glaucoma from fundus images using deep learning techniques
Wan et al. EAD-net: a novel lesion segmentation method in diabetic retinopathy using neural networks
CN111667490B (en) Fundus picture cup optic disc segmentation method
CN109658423B (en) Automatic optic disk cup segmentation method for color fundus picture
CN109919938B (en) Method for obtaining optic disc segmentation atlas of glaucoma
Bhatkalkar et al. Automated fundus image quality assessment and segmentation of optic disc using convolutional neural networks
Zhang et al. Joint optimization of CycleGAN and CNN classifier for detection and localization of retinal pathologies on color fundus photographs
CN108985372A (en) A kind of medical image classification method routed based on capsule theory and PLSA
CN111833334A (en) Fundus image feature processing and analyzing method based on twin network architecture
Shyamalee et al. Glaucoma detection with retinal fundus images using segmentation and classification
Sangeethaa Presumptive discerning of the severity level of glaucoma through clinical fundus images using hybrid PolyNet
CN113610842A (en) OCT image retina detachment and splitting automatic segmentation method based on CAS-Net
CN110472673A (en) Parameter regulation means, method for processing fundus images, device, medium and equipment
CN111784687A (en) Glaucoma fundus image detection method based on deep learning
KR102281968B1 (en) Deep learning architecture system for automatic interpretation of fundus image and method for automatic interpretation of fundus image thereof
CN117315258A (en) Lightweight retinal vessel segmentation method based on graph convolution network and partial convolution

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