CN111563884A - Neural network-based fundus disease identification method, computer device, and medium - Google Patents
Neural network-based fundus disease identification method, computer device, and medium Download PDFInfo
- Publication number
- CN111563884A CN111563884A CN202010339678.8A CN202010339678A CN111563884A CN 111563884 A CN111563884 A CN 111563884A CN 202010339678 A CN202010339678 A CN 202010339678A CN 111563884 A CN111563884 A CN 111563884A
- Authority
- CN
- China
- Prior art keywords
- fundus
- images
- layer
- model
- pooling layer
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Abstract
The invention discloses a method, computer equipment and medium for identifying fundus diseases based on a neural network, wherein the method comprises the following steps: acquiring a plurality of groups of fundus images to be identified, and preprocessing each group of fundus images; inputting the preprocessed fundus image into a fundus image recognition model trained in advance, so as to obtain the fundus disease condition of the patient; the fundus image recognition model is a double-roadbed model which is used for splicing two VGG16 classification model results and outputting fundus disease conditions through a multi-path full-connection layer. The invention carries out the identification of the fundus diseases on the fundus images by splicing the two trained classification models VGG16 to form a new two-way base model, can quickly and efficiently identify the fundus lesions according to the fundus images of the two eyes of a patient, and has strong robustness.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to a method, computer equipment and a medium for recognizing eyeground diseases based on a neural network.
Background
With the rapid development of Chinese economy, the living standard and thought of people are continuously improved. But also cannot ignore the social problems that are facing: the incidence of fundus diseases, particularly myopia, tends to increase year by year, and the shortage of medical resources is opposite to the trend. Early detection of common ocular diseases is very difficult because symptoms are rarely seen in the early stages of the disease. For example, people with longer-term diabetes have a higher chance of developing diabetic retinopathy and diabetic macular edema. Early signs of diabetic eye disease are microaneurysms, which are microscopic and difficult to detect. The macula protects a person's central vision. Macular fluid can distort vision. Age-related macular degeneration is asymptomatic in the early stages. Cataracts are also common in the elderly. It can reduce human vision. Glaucoma refers to a group of diseases that result in damage to the optic nerve head, which is irreversible. In addition, hypertension can alter the morphological structure of blood vessels, e.g., diameter changes and tortuosity changes. It can cause cardiovascular and cerebrovascular diseases such as apoplexy and heart disease. Finally, vision loss is also at high risk in myopic populations due to progressive retinal pigment epithelium thinning and attenuation. Early detection of these diseases can prevent visual impairment and other problems. The image recognition technology can assist an ophthalmologist in diagnosis and help the ophthalmologist to complete fussy fundus image screening work.
The prior art methods comprise an image identification method based on wavelet moments, an image identification method based on fractal characteristics and the like; however, the prior art method has poor robustness and inaccurate classification.
In view of the above, it is desirable to provide a fundus disease identification method capable of quickly and efficiently identifying fundus lesions from fundus images of both eyes of a patient and having high robustness.
Disclosure of Invention
In order to solve the technical problem, the invention adopts the technical scheme that a fundus disease identification method based on a neural network is provided, and the fundus disease identification method based on the neural network comprises the following steps:
acquiring a plurality of groups of fundus images to be identified, and preprocessing each group of fundus images;
inputting the preprocessed fundus image into a fundus image recognition model trained in advance, so as to obtain the fundus disease condition of the patient;
the fundus image recognition model is a double-roadbed model which is used for splicing two VGG16 classification model results and outputting fundus disease conditions through a multi-path full-connection layer.
In the above method, the double roadbed model structure specifically comprises:
the VGG16 basic model comprises two VGG16 basic models with the same structure, wherein the structures are respectively as follows:
input layer input _1_0 → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → pooling layer Conv2D → pooling layer MaxPooling2D → average pooling layer globalaverapoling 2D;
input layer input _2_1 → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → pooling layer Conv2D → pooling layer MaxPooling2D → average pooling layer globalaverapoling 2D;
the two average pooling layers globalaveragePooling2D are simultaneously connected to the splice layer Concatenate, and the splice layer Concatenate is finally connected with 8 full-connection layers Dense for outputting 8 cases of the middle-eye and bottom-eye diseases respectively.
In the above method, the fundus image recognition model is trained by:
acquiring a training set, a verification set and a test set formed by a plurality of groups of fundus images, and preprocessing each group of fundus images in the training set, the verification set and the test set;
respectively inputting two images in each group of images in the preprocessed training set and verification set to two input layers of the initial fundus image recognition model to train the model, and finishing training of the initial fundus image recognition model when the score of the verification set is not increased any more and the training condition is reached; and respectively inputting two images in each group of images of the preprocessed test set into the trained model, and obtaining a final recognition result through multiple convolution calculation, pooling calculation and activation calculation by a forward propagation method.
In the above method, the preprocessing the fundus image includes:
removing row black borders and column black borders on each group of fundus images by using cv2 and numpy programs;
and performing conventional enhancement processing on each group of eye fundus images after the black removing processing, and performing CLAHE enhancement processing on each group of eye fundus images after the enhancement processing by using a CLAHE algorithm.
In the method, image sharpening processing is further performed on each group of fundus images subjected to CLAHE enhancement processing by an image sharpening method.
In the method, a sigmoid function is adopted as a full connection layer Dense activation function, and binary cross entropy loss and difference loss are adopted as a loss function.
The invention also provides computer equipment which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and is characterized in that the processor executes the computer program to realize the fundus disease identification method based on the neural network.
The present invention also provides a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the neural network-based fundus disease identification method as described above.
The invention carries out the identification of the fundus diseases on the fundus images by splicing the two trained classification models VGG16 to form a new two-way base model, can quickly and efficiently identify the fundus lesions according to the fundus images of the two eyes of a patient, and has strong robustness.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a method provided by the present invention;
FIG. 2 is a block diagram of a two-way base model architecture provided by the present invention;
FIG. 3 is a flow chart of an initial fundus image recognition model training provided by the present invention;
FIG. 4 is an original binocular fundus image to be recognized in the method provided by the present invention, with fig. (a) being a left eye and fig. (b) being a right eye;
FIG. 5 is a binocular fundus image after the black border processing of FIG. 4 according to the present invention;
FIG. 6 shows fundus images of both eyes after conventional enhancement processing of FIG. 5, wherein (c) shows the left eye and (d) shows the right eye;
FIG. 7 is a binocular fundus image after CLAHE enhancement processing is performed on FIG. 6 according to the present invention;
FIG. 8 is a histogram after threshold correction provided by the present invention;
FIG. 9 is a diagram illustrating a fundus image of both eyes after sharpening process is performed on FIG. 7 according to the present invention;
FIG. 10 is a binocular fundus image of FIG. 4 after being processed by black edge removal, conventional enhancement, CLAHE enhancement and image sharpening in the present invention;
fig. 11 is a block diagram schematically illustrating the structure of a computer device according to the present invention.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the following embodiments, and it should be understood that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention is described in detail below with reference to specific embodiments and the accompanying drawings.
As shown in fig. 1, the present invention provides a method for identifying fundus diseases based on a neural network, comprising the steps of:
s1, acquiring a plurality of groups of fundus images to be identified, and preprocessing each group of fundus images;
s2, inputting the preprocessed fundus image into a fundus image recognition model trained in advance, so as to obtain the fundus disease condition of the patient; the fundus image recognition model is a double-roadbed model which is used for splicing two VGG16 classification model results and outputting fundus disease conditions through a multi-path full-connection layer.
In this embodiment, the two-way basic model is specifically shown in fig. 2, and includes two VGG16 classification models with the same structure, where the structures are:
input layer input _1_0 → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → pooling layer Conv2D → pooling layer MaxPooling2D → average pooling layer globalaverapoling 2D;
input layer input _2_1 → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → pooling layer Conv2D → pooling layer MaxPooling2D → average pooling layer globalaverapoling 2D;
the two average pooling layers globalaveragePooling2D are simultaneously connected to the splice layer Concatenate, and the splice layer Concatenate is finally connected to 8 full-connection layers Dense for outputting 8 cases of the fundus diseases.
The model is an end-to-end model, and the structure is shown in fig. 2, where Input _1_0 and Input _2_1 are image Input layers, which can be respectively Input as left/right eye images, Conv2D is convolution operation, MaxPooling2D is maximum pooling operation, globavaveragepooling 2D is average pooling operation on channels, Concatenate is splicing operation on channels, and density is output from fully connected layers Dense _ 1-8.
The binocular fundus information of the patient can be obtained through two-way input, and the basic model can prevent model overfitting by adopting a VGG16 classification model.
In this embodiment, the 8-way dense _1-8 outputs the information of fundus diseases that can be identified as normal (N), diabetes (D), glaucoma (G), cataract (C), amd (a), hypertension (H), myopia (M) and other diseases/abnormalities (O) according to the requirements, and the output conditions are divided into several types, for example, one patient may have multiple fundus diseases, and at this time, a multi-way result will be output, and if the model finally identifies that the fundus disease condition is diabetes (D), glaucoma (G) and other diseases/abnormalities (O), dense2, dense3 and dense8 will output corresponding results.
In this embodiment, if a fundus in one of the fundus images is recognized as having a disease, the patient has a disease.
In the embodiment, the trained two classification models VGG16 are spliced to form a new two-way base model to identify the fundus diseases of the fundus images, so that fundus lesions can be identified quickly and efficiently according to the fundus images of the two eyes of a patient, and the robustness is high.
As shown in fig. 3, in the present embodiment, the fundus image recognition model is obtained by training in the following manner:
in the present embodiment, a data set (a plurality of sets of fundus images) including 4500 pairs of fundus images, in which 500 pairs of fundus images are randomly extracted as a test set, 500 pairs of fundus images are randomly extracted as a verification set in images not including the test set, and all the remaining images are extracted as a training set, is divided into training set, verification set, and test set 3 portions.
Training is carried out by using images of a training set and a verification set, wherein the images of the training set and the verification set are subjected to black edge removing, conventional enhancement, CLAHE enhancement and image sharpening, two classification models VGG16 are trained by using an Adam optimized gradient descent algorithm, and the training is finished when the training condition is reached until the score of the verification set is not increased any more; and then, after the test collection is subjected to black edge removal, CLAHE enhancement and image sharpening, inputting the test collection into a model, and obtaining a final identification result through a forward propagation method and multiple times of convolution calculation, pooling calculation and activation calculation.
A1, acquiring a training set, a verification set and a test set formed by a plurality of groups of fundus images, and preprocessing each group of fundus images in the training set, the verification set and the test set;
the data sets of the present embodiment are "real" patient information collected from different hospitals/medical centers in the country. In these mechanisms, fundus images are captured by various cameras on the market, such as Canon, Zeiss, and Kowa, thus resulting in a wide variety of image resolutions.
A2, respectively inputting two images in each group of images in the preprocessed training set and verification set to two input layers of the initial fundus image recognition model to train the model, and finishing training of the initial fundus image recognition model when the score of the verification set is not increased any more and reaches a training condition; and respectively inputting two images in each group of images of the preprocessed test set into the trained model, and obtaining a final recognition result through multiple convolution calculation, pooling calculation and activation calculation by a forward propagation method.
And obtaining the trained fundus image recognition model.
In the present embodiment, the method of preprocessing the fundus images (including the fundus image to be recognized in step S1 and the fundus images in the training set of step a 1) is specifically as follows:
since there is a "black edge" in the "original" image, the black edge means an array of "height × 1 × 3" or "1 × width × 3" and all the values are 0, the black edge is useless information, which would reduce the efficiency of the model, and therefore the black edge needs to be removed, in this embodiment, by using cv2 and numpy program to perform dimensionality reduction and summation on the color channel and the width channel of the image, the obtained index from the top to the bottom first non-0 to the last non-0 is the height to be preserved, that is, the black edge of the row is removed. The method for removing the column black edge is the same as the method and is not repeated; as shown in fig. 4 and fig. 5, the original fundus image to be processed and the fundus images of both eyes after the black border is removed using the method are respectively shown.
In order to improve the robustness of the model in the embodiment, the embodiment performs enhancement processing on the binocular fundus image from which the black edge is removed, and a conventional enhancement method can be adopted, including: translation, rotation, flipping, clipping, zooming. As shown in fig. 6, a left eye fundus image (fig. a) and a right eye fundus image (fig. b) after conventional enhancement are respectively.
In order to enhance the contrast of the fundus image, the embodiment further processes the conventionally enhanced fundus image by the CLAHE algorithm, as shown in fig. 7, which is a schematic diagram of the fundus image enhanced by the CLAHE algorithm, and the specific processing calculation is as follows:
processing the fundus image using a CLAHE algorithm, as in equations (1) - (4):
f(i,j)＝mx(i,j)+G(i,j)[x(i,j)-mx(i,j)](4)
wherein x (k, l) is the gray value of the pixel point (k, l); m isx(i, j) is a local area gray average value with a window size of (2n +1) × (2n +1) centered on (i, j);
is the local area variance; g (i, j) is a contrast enhancement coefficient, defined as the global mean/local area standard deviation of the image; f (i, j) is the gray value of the pixel point (i, j) after CLAHE enhancement; in order to limit the over-amplification noise, the histogram of x (i, j) (the parameter is the gray value of (i, j) point in the rectangular coordinate system, and the range is 0-255) is transformed as shown in fig. 8, and the number of pixels with the frequency higher than the threshold is reduced to the threshold.
In order to further improve the characteristics of the fundus image, the embodiment sharpens the fundus image processed by the CLAHE method by using a sharpening method, and the sharpened fundus image is as shown in fig. 9; the sharpening method is specifically as follows:
enhanced(i,j)＝4[x(i,j)-GaussianBlur(i,j)]+128 (5)
wherein x (i, j) is the gray value of the pixel point (i, j); GaussianBlur (i, j) is the Gaussian blur value of the pixel point (i, j), and is not described again; enhanced (i, j) is a value of the pixel point (i, j) after sharpening.
As shown in fig. 10, left eye fundus images and right eye fundus images of fundus images subjected to the above-described black edge removal, normal enhancement, CLAHE enhancement, and image sharpening are shown.
In this embodiment, the selection of the training parameters of the initial fundus image recognition model is specifically as follows:
because the model output is 8 binary classification problems, the activation function of the embodiment adopts a sigmoid function as formula (6), and the loss function adopts a binary cross entropy loss as formula (7); further, in order to distinguish the normal fundus condition from 7 fundus diseases, the difference loss as in the formula (8) was also added.
In the formula (6), x is a pre-activation numerical value, and sigmoid is an activated output;
y in the formula (7) is a true value,
is the prediction value binary _ cross entropy _ loss;
in the formula (8), the reaction mixture is,
is 1 normal scoreThe value of the class prediction is used,
is the maximum of the predictive values of 7 disease classifications, and difference _ loss is the value of difference loss.
The model of the embodiment is initialized by adopting imagenet transfer learning, and the optimization method adopts Adam, as shown in formula (9):
in the formula, mtIs an updated biased first moment estimate; v. oftIs an updated biased second moment estimate;
is a corrected unbiased first moment estimate;
is a corrected unbiased second moment estimate;
in this example, α is 0.001 (learning rate), β1＝0.9，β2＝0.999，＝10-8Is a hyper-parameter; gtIs the gradient at time t; thetatIs the updated parameter.
The method is described below by way of specific examples.
In the case, by using the method to identify 500 pairs of eye fundus images, using Titan Xp GPU with batch of 4, only 3.75 minutes are needed to identify the images to obtain an identification result, and the average value of AUC, KAPPA and F1 is as high as 0.85; wherein the diagnosis time is timed by a time module in the system, the total time T predicted by 500 pairs of eye fundus images is counted for i times, and the average value T _uis taken for the total timeavg＝(T1+T2+...+Ti) I, finally calculating the average time T of the single sheet as T ═ T \ u by using the total timeavg/500＝30ms；
Under the condition of the same speed, if the image recognition method using the wavelet moment or the image recognition method based on the fractal feature is adopted, the same score cannot be achieved; also if the examination is performed by an experienced ophthalmologist, it takes 1.38 hours to complete the examination, assuming that the examination speed is 5 s/sheet, and even more, it is difficult for the expert to achieve the examination speed. The rapidness, high efficiency and strong robustness of the method are demonstrated.
Therefore, the disease condition in the fundus image identified by the method far exceeds the average diagnosis speed of a doctor, and the model not only can solve the problem of complicated and repeated diagnosis work of the doctor and enable the doctor to put energy on more important things, but also can provide reasonable reference opinions for the doctor.
As shown in fig. 11, the present invention also provides a computer device including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the neural network-based fundus disease identification method in the above-described embodiment when executing the computer program.
The present invention also provides a computer-readable storage medium on which a computer program is stored, the computer program, when executed by a processor, implementing the recognition model training method in the above-described embodiments, or the computer program, when executed by the processor, implementing the neural network-based fundus disease recognition method in the above-described embodiments.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The present invention is not limited to the above-mentioned preferred embodiments, and any structural changes made under the teaching of the present invention shall fall within the protection scope of the present invention, which has the same or similar technical solutions as the present invention.
Claims (8)
1. The fundus disease identification method based on the neural network is characterized by comprising the following steps of:
acquiring a plurality of groups of fundus images to be identified, and preprocessing each group of fundus images;
inputting the preprocessed fundus image into a fundus image recognition model trained in advance, so as to obtain the fundus disease condition of the patient;
wherein, the fundus image recognition model is as follows: and splicing the two VGG16 classification models and outputting a double-subgrade model of the fundus disease condition through a multi-path full-connection layer.
2. The method for identifying fundus diseases based on a neural network according to claim 1, wherein said dual roadbed model structure is specifically:
the VGG16 basic model comprises two VGG16 basic models with the same structure, wherein the structures are respectively as follows:
input layer input _1_0 → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → pooling layer Conv2D → pooling layer MaxPooling2D → average pooling layer globalaverapoling 2D;
input layer input _2_1 → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 2 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → convolution layer Conv2D × 3 → pooling layer MaxPooling2D → pooling layer Conv2D → pooling layer MaxPooling2D → average pooling layer globalaverapoling 2D;
the two average pooling layers globalaveragePooling2D are simultaneously connected to the splice layer Concatenate, and the splice layer Concatenate is finally connected with 8 full-connection layers Dense for outputting 8 cases of the middle-eye and bottom-eye diseases respectively.
3. The neural-network-based fundus disease recognition method according to claim 1 or 2, wherein the fundus image recognition model is trained by:
acquiring a training set, a verification set and a test set formed by a plurality of groups of fundus images, and preprocessing each group of fundus images in the training set, the verification set and the test set;
respectively inputting two images in each group of images in the preprocessed training set and verification set to two input layers of the initial fundus image recognition model to train the model, and finishing training of the initial fundus image recognition model when the score of the verification set is not increased any more and the training condition is reached; and respectively inputting two images in each group of images of the preprocessed test set into the trained model, and obtaining a final recognition result through multiple convolution calculation, pooling calculation and activation calculation by a forward propagation method.
4. The neural-network-based fundus disease identification method according to claim 1 or 2, wherein said preprocessing the fundus image includes:
removing row black borders and column black borders on each group of fundus images by using cv2 and numpy programs;
and performing conventional enhancement processing on each group of eye fundus images after the black removing processing, and performing CLAHE enhancement processing on each group of eye fundus images after the enhancement processing by using a CLAHE algorithm.
5. The neural-network-based fundus disease identification method of claim 4, further comprising
And carrying out image sharpening on each group of fundus images subjected to CLAHE enhancement processing by using an image sharpening method.
6. The neural-network-based fundus disease identification method according to claim 3, wherein said
The full-connection layer Dense activation function adopts a sigmoid function, and the loss function adopts binary cross entropy loss and difference loss.
7. Computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the neural network based fundus disease identification method according to any one of claims 1 to 6 when executing the computer program.
8. A computer-readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the neural network-based fundus disease identification method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010339678.8A CN111563884A (en) | 2020-04-26 | 2020-04-26 | Neural network-based fundus disease identification method, computer device, and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010339678.8A CN111563884A (en) | 2020-04-26 | 2020-04-26 | Neural network-based fundus disease identification method, computer device, and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111563884A true CN111563884A (en) | 2020-08-21 |
Family
ID=72074388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010339678.8A Pending CN111563884A (en) | 2020-04-26 | 2020-04-26 | Neural network-based fundus disease identification method, computer device, and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111563884A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112580530A (en) * | 2020-12-22 | 2021-03-30 | 泉州装备制造研究所 | Identity recognition method based on fundus images |
| CN113158863A (en) * | 2021-04-13 | 2021-07-23 | 同济大学 | Abnormal fundus photo identification method |
| CN113222061A (en) * | 2021-05-31 | 2021-08-06 | 北京理工大学 | MRI image classification method based on two-way small sample learning |
| CN116433644A (en) * | 2023-04-22 | 2023-07-14 | 深圳市江机实业有限公司 | Eye image dynamic diagnosis method based on recognition model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019024568A1 (en) * | 2017-08-02 | 2019-02-07 | 上海市第六人民医院 | Ocular fundus image processing method and apparatus, computer device, and storage medium |
| CN110400288A (en) * | 2019-06-18 | 2019-11-01 | 中南民族大学 | A kind of sugar of fusion eyes feature nets sick recognition methods and device |
-
2020
- 2020-04-26 CN CN202010339678.8A patent/CN111563884A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019024568A1 (en) * | 2017-08-02 | 2019-02-07 | 上海市第六人民医院 | Ocular fundus image processing method and apparatus, computer device, and storage medium |
| CN110400288A (en) * | 2019-06-18 | 2019-11-01 | 中南民族大学 | A kind of sugar of fusion eyes feature nets sick recognition methods and device |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112580530A (en) * | 2020-12-22 | 2021-03-30 | 泉州装备制造研究所 | Identity recognition method based on fundus images |
| CN113158863A (en) * | 2021-04-13 | 2021-07-23 | 同济大学 | Abnormal fundus photo identification method |
| CN113222061A (en) * | 2021-05-31 | 2021-08-06 | 北京理工大学 | MRI image classification method based on two-way small sample learning |
| CN113222061B (en) * | 2021-05-31 | 2022-12-09 | 北京理工大学 | MRI image classification method based on two-way small sample learning |
| CN116433644A (en) * | 2023-04-22 | 2023-07-14 | 深圳市江机实业有限公司 | Eye image dynamic diagnosis method based on recognition model |
| CN116433644B (en) * | 2023-04-22 | 2024-03-08 | 深圳市江机实业有限公司 | Eye image dynamic diagnosis method based on recognition model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kwasigroch et al. | Deep CNN based decision support system for detection and assessing the stage of diabetic retinopathy | |
| CN111563884A (en) | Neural network-based fundus disease identification method, computer device, and medium | |
| Khan et al. | Cataract detection using convolutional neural network with VGG-19 model | |
| Arora et al. | Deep neural network for diabetic retinopathy detection | |
| CN109726743B (en) | Retina OCT image classification method based on three-dimensional convolutional neural network | |
| CN113889267A (en) | Method for constructing diabetes diagnosis model based on eye image recognition and electronic equipment | |
| Al-Bander et al. | A novel choroid segmentation method for retinal diagnosis using deep learning | |
| Ştefan et al. | A review of glaucoma detection from digital fundus images using machine learning techniques | |
| Phridviraj et al. | A bi-directional Long Short-Term Memory-based Diabetic Retinopathy detection model using retinal fundus images | |
| CN117058676B (en) | Blood vessel segmentation method, device and system based on fundus examination image | |
| Giancardo | Automated fundus images analysis techniques to screen retinal diseases in diabetic patients | |
| Kamal et al. | A comprehensive review on the diabetic retinopathy, glaucoma and strabismus detection techniques based on machine learning and deep learning | |
| CN113610842A (en) | OCT image retina detachment and splitting automatic segmentation method based on CAS-Net | |
| Gour et al. | Blood vessel segmentation using hybrid median filtering and morphological transformation | |
| Esfahani et al. | Automatic Classification of Macular Diseases from OCT Images Using CNN Guided with Edge Convolutional Layer | |
| Poletti et al. | Automatic nerve tracking in confocal images of corneal subbasal epithelium | |
| Triyadi et al. | Deep learning in image classification using vgg-19 and residual networks for cataract detection | |
| CN116452571A (en) | Image recognition method based on deep neural network | |
| Roy et al. | Retnet: Retinal disease detection using convolutional neural network | |
| Yijie et al. | Improved U-net fundus image segmentation method | |
| Bijam et al. | A review on detection of diabetic retinopathy using deep learning and transfer learning based strategies | |
| Hatode et al. | Evolution and Testimony of Deep Learning Algorithm for Diabetic Retinopathy Detection | |
| Dufour et al. | Classification of drusen positions in optical coherence tomography data from patients with age-related macular degeneration | |
| Saleh et al. | Early Diabetic Retinopathy Detection Using Convolution Neural Network | |
| Ghorab et al. | Computer-Based Detection of Glaucoma Using Fundus Image Processing |
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 |