CN110210570A - The more classification methods of diabetic retinopathy image based on deep learning - Google Patents
The more classification methods of diabetic retinopathy image based on deep learning Download PDFInfo
- Publication number
- CN110210570A CN110210570A CN201910495481.0A CN201910495481A CN110210570A CN 110210570 A CN110210570 A CN 110210570A CN 201910495481 A CN201910495481 A CN 201910495481A CN 110210570 A CN110210570 A CN 110210570A
- Authority
- CN
- China
- Prior art keywords
- image
- diabetic retinopathy
- training
- deep learning
- convolutional 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
- 206010012689 Diabetic retinopathy Diseases 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 72
- 238000012546 transfer Methods 0.000 claims abstract description 20
- 238000005070 sampling Methods 0.000 claims description 24
- 201000010099 disease Diseases 0.000 claims description 12
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 12
- 208000017442 Retinal disease Diseases 0.000 claims description 11
- 206010038923 Retinopathy Diseases 0.000 claims description 11
- 206010012601 diabetes mellitus Diseases 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 5
- 238000013519 translation Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000006467 substitution reaction Methods 0.000 claims description 3
- 238000003475 lamination Methods 0.000 claims description 2
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000012545 processing Methods 0.000 description 4
- 210000001525 retina Anatomy 0.000 description 4
- 230000003321 amplification Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 230000003902 lesion Effects 0.000 description 3
- 238000003199 nucleic acid amplification method Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002062 proliferating effect Effects 0.000 description 2
- 230000004256 retinal image Effects 0.000 description 2
- 238000005303 weighing Methods 0.000 description 2
- 206010018473 Glycosuria Diseases 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 210000002700 urine Anatomy 0.000 description 1
Classifications
-
- 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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Medical Informatics (AREA)
- Evolutionary Biology (AREA)
- Databases & Information Systems (AREA)
- Primary Health Care (AREA)
- General Health & Medical Sciences (AREA)
- Epidemiology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
Present invention discloses the more classification methods of diabetic retinopathy image based on deep learning, include the following steps: firstly, obtaining raw data set, and raw data set is pre-processed；Then pretreated data set is subjected to categorical data balance；Later, the data set after categorical data being balanced carries out transfer training, obtains the more disaggregated models of diabetic retinopathy image；Finally, the sample to be tested input more disaggregated models of diabetic retinopathy image are predicted, and complete the classification to diabetic retinopathy image.Method provided by the invention has the advantages that easy to use, in addition, the present invention carries out transfer training by data set after balancing categorical data, improves classification accuracy.
Description
Technical field
The present invention relates to a kind of image classification method more particularly to a kind of diabetic retinopathy based on deep learning
The more classification methods of image belong to medical image identification technology field.
Background technique
In medical system at this stage, tend to artificial detection mostly, the processing of diabetic retina image also not example
Outside, diabetic retina image lesion is divided into five kinds, respectively aglycosuria characteristic of disease retinopathy, slight sugar according to menace level
Urinate characteristic of disease retinopathy, medium diabetes mellitus retinopathy, severe diabetic retinopathy change and Proliferative diabetes
Property retinopathy.
Traditional diabetic retina image processing method is that system is carried out section, in the different stages using difference
Processing technique reach target, still, no matter be based on artificial experience in which or which kind of processing technique method in stage and set
Feature is counted, i.e., characteristics of image, such as histograms of oriented gradients, scale invariant feature conversion are extracted according to the particularity of particular problem
Deng.Although these features embody the feature of object to a certain extent, they are only extracted low-level image feature, such as scheme
Edge feature, gray feature of picture etc., in addition retinal images self structure is complicated, easy and various lesion cross influences, then
In addition complicated background changing influences, so that the method for processing retinal images is complicated, generalization is poor；In addition, this detection side
Formula depends on priori knowledge, is especially the need for the ophthalmologist, oculist of experience.Therefore, whole process not only consumes a large amount of people
The Clinics and Practices of power, material resources and financial resources, and inefficiency, retinopathy are greatly limited, and finally make image point
There is very big error in class testing result, influences more classification results.
In conclusion how to reduce the error of image classification testing result, the accuracy of classification results is improved, just becomes this
Field technical staff's urgent problem to be solved.
Summary of the invention
The purpose of the present invention is to solve the drawbacks described above of the prior art, propose a kind of real based on deep learning method
Now to automatic more classification of diabetic retina image.
The technical solution of the invention is as follows:
A kind of more classification methods of diabetic retinopathy image based on deep learning, include the following steps:
S1: obtaining a series of other eye fundus images of five types as raw data set, raw data set pre-processed,
Obtain the picture scale for being suitble to network training；
S2: carrying out categorical data balance based on data set pretreated in step S1, and after categorical data is balanced
Data set is divided into training set and test set, and the mode of the categorical data balance includes for keeping the training set number flat
The up-sampling and down-sampling of weighing apparatus；
S3: the data set after categorical data balance in step S2 is subjected to transfer training, the transfer training includes first
Secondary transfer training and second of transfer training, first time transfer training are that small size training set image is input in network to carry out
Training generates characteristic pattern, wherein training network shallow-layer uses VGG16 network architecture parameters, and deep layer network parameter is according to training set
Image carries out first time fine tuning；Second of transfer training is will to finely tune obtained model for the first time to migrate to large scale training set figure
It is finely tuned as carrying out second, obtains the more disaggregated models of diabetic retinopathy image；
S4: sample to be tested is inputted into the more disaggregated models of the diabetic retinopathy image in the step S3 and is carried out in advance
It surveys, and completes the classification to diabetic retinopathy image.
Preferably, the pretreatment in the step S1 includes gaussian filtering process and down-sampled；
Gaussian filtering is with each of convolution scan image pixel, and pixel adds in the neighborhood determined with convolution
The value of weight average gray value substitution convolution central pixel point；
Down-sampled is to reduce sampling number, is then in the N*M if down-sampled coefficient is k for the image of a width N*M
Image graph in each row and column every k point take point composition piece image so that image meets network training scale.
Preferably, network training picture scale is 256*256 in the step S1.
Preferably, the up-sampling in the step S2 is the sampling to the lower progress of the other training set quantity of five types, institute
Stating the down-sampling in step S2 is the sampling to the higher progress of the other training set quantity of five types.
Preferably, the mode of the categorical data balance in the step S2 further include translation, it is random stretch, rotation and its
In conjunction with.
Preferably, the more disaggregated models of diabetic retinopathy image successively include input layer, convolutional layer, convolution
Layer, maximum pond layer, convolutional layer, convolutional layer, maximum pond layer, convolutional layer, convolutional layer, convolutional layer, maximum pond layer, convolution
Layer, convolutional layer, convolutional layer, maximum pond layer, the first full articulamentum, the second full articulamentum, the full articulamentum of third.
Preferably, the more disaggregated models of diabetic retinopathy image further include for preventing over-fitting and increasing non-
Linear dropout layer, described dropout layers is arranged after the full articulamentum of third.
Preferably, the side that each maximum pond layer is connect with the convolutional layer of its two sides by part and weight is shared
Formula is connected；Each convolutional layer, the first full articulamentum, the second full articulamentum and the full articulamentum of the third
It is provided with the ReLU activation primitive for guaranteeing Neural Network Based Nonlinear afterwards.
Preferably, the output of the full articulamentum of the third uses 5 neurons and softmax function to five types of disease
Do not classify.
Preferably, the algorithm of the more disaggregated models of diabetic retinopathy image includes:
Training set is made of m marked samples: { (x(1), y(1)) ..., (x(m), y(m)), wherein input feature vector is x(i)∈Rn+1, classification results y(i)∈ { 1,2,3,4,5 }, then function hθ(x) form is as follows:
Cost function are as follows:
For the minimization problem of J (θ), it is as follows to obtain gradient formula by derivation using the optimization algorithm of iteration:Iteration requires each time
It is updated as follows:By minimizing J (θ), diabetes view is realized
The more disaggregated models of retinopathy image.
The present invention provides a kind of more classification methods of diabetic retinopathy image based on deep learning, advantage
Are as follows: method provided by the invention is not needed by any prior information using easy, is done simple image after input data and is located in advance
Reason, the image for then obtaining pretreatment are trained as sample by the more disaggregated models of diabetic retinopathy image automatically,
Image more classification automatically finally can be completed；In addition, the present invention carries out migration instruction by the data set after balancing categorical data
Practice, and predicted by the more disaggregated models of diabetic retinopathy image, improves classification accuracy.
Just attached drawing in conjunction with the embodiments below, the embodiment of the present invention is described in further detail, so that of the invention
Technical solution is more readily understood, grasps.
Detailed description of the invention
Fig. 1 is the flow chart of the more classification methods of diabetic retinopathy image based on deep learning.
Specific embodiment
A kind of more classification methods of diabetic retinopathy image based on deep learning, as shown in Figure 1, including following step
It is rapid:
S1: obtaining a series of other eye fundus images of five types as raw data set, raw data set pre-processed,
Obtain the picture scale for being suitble to network training；
In the inventive solutions, raw data set is the data from kaggle sugar net high-definition picture
Collection, resolution ratio is about 3500*3000, includes altogether 35126 images, is divided into five classes, including sugar-free by the severity of disease
Open, mild diabetes retinopathy 2443 is opened, medium diabetes mellitus retinopathy by characteristic of disease retinopathy 25810 for urine
5292, severe diabetic retinopathy change 873, proliferating diabetic retinopathy change 708.To raw data set
Carrying out pretreated situation includes following four: the first is that resolution ratio is excessive and scale is not suitable for network training input；
The noise image for being for second this meaningless information of picture black surround for including in image and including in data set；The third is
The shading value difference of image is larger；4th kind is that the serious imbalance of amount of images leads to not be sent directly into mould in five kinds of classifications
Type.
Pretreatment includes gaussian filtering process and down-sampled；Wherein, gaussian filtering is exactly weighted entire image flat
Equal process, the value of each pixel obtain after being all weighted averagely by other pixel point values in itself and neighborhood.
The concrete operations of gaussian filtering are: with each of convolution scan image pixel, pixel in the neighborhood that is determined with convolution
Weighted average gray value substitution convolution central pixel point value；The black surround in image can be cut off using gaussian filtering process,
And eliminate the influence of noises such as illumination, smoothed image.Down-sampled is that sampling number is reduced, for the image of a width N*M, if down-sampled
Coefficient is k, then be in the image graph of the N*M each row and column every k point take point composition piece image so that image
Meet network training scale, and network training picture scale is 256*256.
S2: carrying out categorical data balance based on data set pretreated in step S1, and after categorical data is balanced
Data set is divided into training set and test set, and the mode of the categorical data balance includes for keeping the training set number flat
The up-sampling and down-sampling of weighing apparatus further include translation, stretch, rotate and its combine, to five kinds of classifications at random using translation, rotation
Etc. data amplification method generate new data, to prevent model over-fitting；Up-sampling is lower to the other training set quantity of five types
The sampling of progress, down-sampling are the samplings to the higher progress of the other training set quantity of five types.In the present embodiment, up-sampling will
The less image of sample size copies to 10000, and the down-sampling classification more to sample size carries out random sampling 10000
, five classifications share 50000 images after classification balance.
S3: the data set after categorical data balance in step S2 is subjected to transfer training, the transfer training includes first
Secondary transfer training and second of transfer training, first time transfer training are that small size training set image is input in network to carry out
Training generates characteristic pattern, wherein training network shallow-layer uses VGG16 network architecture parameters, and deep layer network parameter is according to training set
Image carries out first time fine tuning；Second of transfer training is will to finely tune obtained model for the first time to migrate to large scale training set figure
It is finely tuned as carrying out second, obtains the more disaggregated models of diabetic retinopathy image；
In the present embodiment, first to prevent model over-fitting, the data of training set are generated by Data Generator, to five kinds
Classification generates new data using data amplification methods such as translation, rotations at random.It is defeated that 256*256 is converted by training set picture size
Enter and be trained into network, generate characteristic pattern, training network shallow-layer uses VGG16 network architecture parameters, deep layer network parameter
Be finely adjusted according to training data, by obtained disaggregated model again with picture size be the 512*512 model finely tuned of input into
Row is finely tuned again, finally obtains the more disaggregated models of diabetic retinopathy image.
VGG16 network structure trains model parameter on ImageNet data set, and ImageNet data set is
About the huge data set of natural image disclosed in one.It is low identifying although natural image and disease image are not consistent
It but has points of resemblance when rank semantic information, therefore the parameter for identifying natural image is used to identify disease, by last full connection
Layer changes the full articulamentum of five kinds of classification tasks into, then is finely adjusted, model just can fast convergence, when can not only save model training
Between, and can be reduced the demand to sample size to prevent over-fitting.
The more disaggregated models of diabetic retinopathy image successively include input layer, convolutional layer, convolutional layer, maximum pond
Layer, convolutional layer, convolutional layer, maximum pond layer, convolutional layer, convolutional layer, convolutional layer, maximum pond layer, convolutional layer, convolutional layer, volume
Lamination, maximum pond layer, the first full articulamentum, the second full articulamentum, the full articulamentum of third, further include for prevent over-fitting and
Increase it is dropout layers nonlinear, it is described dropout layers be arranged after the full articulamentum of third.Wherein, each maximum pond layer
It is connected by way of locally connecting and weight is shared with the convolutional layer of its two sides；Each convolutional layer, the first full articulamentum,
The ReLU activation primitive for guaranteeing Neural Network Based Nonlinear is provided with after second full articulamentum and the full articulamentum of third.The
The output of three full articulamentums classifies to five kinds of classifications of disease using 5 neurons and softmax function.
Further, convolutional layer is all the convolution kernel with 3*3, step-length 1, and the first two convolutional layer all uses 64 convolution
Core, two convolutional layers then all use 128 convolution kernels, and three convolutional layers later all use 256 convolution kernels, remaining
Convolutional layer all uses 512 convolution kernels, and the first full articulamentum and the second full articulamentum use 1024 neurons, and third connects entirely
It connects layer and the classification of disease 5 is carried out to the output of full articulamentum using 5 neurons and with softmax function, obtain disease classification knot
Fruit, model use intersection entropy function as loss function and adjust to carry out the backpropagation of parameter.Over-fitting and increasing in order to prevent
Add non-linear, dropout layers are added to after full articulamentum, has ReLU behind each convolutional layer and full articulamentum of network
Activation primitive.
The algorithm of the more disaggregated models of diabetic retinopathy image includes:
Training set is made of m marked samples: { (x(1), y(1)) ..., (x(m), y(m)), wherein input feature vector is x(i)∈Rn+1, classification results y(i)∈ { 1,2,3,4,5 }, then function hθ(x) form is as follows:
Cost function are as follows:
For the minimization problem of J (θ), gradient formula is obtained by derivation using the optimization algorithm of iteration
It is as follows:Often
An iteration requires to be updated as follows:By minimizing J (θ),
Realize the more disaggregated models of diabetic retinopathy image.
S4: sample to be tested is inputted into the more disaggregated models of the diabetic retinopathy image in the step S3 and is carried out in advance
It surveys, and completes the classification to diabetic retinopathy image.
The classification accuracy in the present invention is further illustrated with experimental data below:
Image is opened to every a kind of data pick-up 100 first and is used to assessment models effect, it is remaining to be used as training set.Training
The model of 256*256 image is used to model parameter moving to the more disaggregated models of diabetic retinopathy image from natural image
Image, training 512*512 be that can reach for the model parameter for being applied to low resolution is moved to high-definition picture
Better recognition effect, when the two model trainings data be all generated using Data Generator using the method for data amplification it is new
Data.Training 256*256 image when, due to it is initial when model parameter be suitable for natural image, freeze the first full articulamentum first
Before all layers of parameter trains full articulamentum using the Adam optimizer that parameter is 0.0001, then thaws last six
A convolutional layer finely tunes model parameter using the Adam optimizer that parameter is 0.00002, and test set accuracy rate does not rise three times
Stop training of the model to training set, finally may be implemented to move to the priori knowledge of natural image in sugared net disease, so that
Model accuracy rate reaches 75%；Train the process of the image of 512*512 identical as training 256*256 image later, so that model
It is capable of the lesion of clearer identification microsize, final mask accuracy rate can reach 77%.
The present invention is not needed by any prior information, is done simple image preprocessing after input data, then will be located in advance
Obtained image is managed as sample, is trained by the more disaggregated models of diabetic retinopathy image, finally be can be completed automatically
Image more classification automatically；In addition, the present invention carries out transfer training by the data set after balancing categorical data, and pass through glycosuria
The sick more disaggregated models of retinopathy image are predicted, classification accuracy is improved.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and this
Field technical staff can be designed alternative embodiment without departing from the scope of the appended claims.
Claims (10)
1. a kind of more classification methods of diabetic retinopathy image based on deep learning, which is characterized in that including walking as follows
It is rapid:
S1: a series of other eye fundus images of five types are obtained as raw data set, raw data set is pre-processed, is obtained
It is suitble to the picture scale of network training；
S2: carrying out categorical data balance based on data set pretreated in step S1, and the data after categorical data is balanced
Collection is divided into training set and test set, and the mode of the categorical data balance includes for keeping the training set number to balance
Up-sampling and down-sampling；
S3: the data set after categorical data balance in step S2 is subjected to transfer training, the transfer training includes moving for the first time
Trained and second of transfer training is moved, first time transfer training is that small size training set image is input in network to instruct
Practice, generate characteristic pattern, wherein training network shallow-layer uses VGG16 network architecture parameters, and deep layer network parameter is according to training set figure
As carrying out first time fine tuning；Second of transfer training is will to finely tune obtained model for the first time to migrate to large scale training set image
It carries out second to finely tune, obtains the more disaggregated models of diabetic retinopathy image；
S4: the more disaggregated models of diabetic retinopathy image that sample to be tested inputs in the step S3 are predicted, and
Complete the classification to diabetic retinopathy image.
2. diabetic retinopathy image more classification methods according to claim 1 based on deep learning, feature
Be: the pretreatment in the step S1 includes gaussian filtering process and down-sampled；
Gaussian filtering is with each of convolution scan image pixel, and the weighting of pixel is put down in the neighborhood determined with convolution
The value of equal gray value substitution convolution central pixel point；
Down-sampled is to reduce sampling number, is then the figure in the N*M if down-sampled coefficient is k for the image of a width N*M
As each row and column every k point take point composition piece image in figure, so that image meets network training scale.
3. diabetic retinopathy image more classification methods according to claim 1 based on deep learning, feature
Be: network training picture scale is 256*256 in the step S1.
4. diabetic retinopathy image more classification methods according to claim 1 based on deep learning, feature
Be: the up-sampling in the step S2 is the sampling to the lower progress of the other training set quantity of five types, in the step S2
Down-sampling be the sampling to the higher progress of the other training set quantity of five types.
5. diabetic retinopathy image more classification methods according to claim 4 based on deep learning, feature
Be: the mode of the categorical data balance in the step S2 further includes translation, stretches, rotates and its combine.
6. diabetic retinopathy image more classification methods according to claim 1 based on deep learning, feature
Be: the more disaggregated models of diabetic retinopathy image successively include input layer, convolutional layer, convolutional layer, maximum pond
Layer, convolutional layer, convolutional layer, maximum pond layer, convolutional layer, convolutional layer, convolutional layer, maximum pond layer, convolutional layer, convolutional layer, volume
Lamination, maximum pond layer, the first full articulamentum, the second full articulamentum, the full articulamentum of third.
7. diabetic retinopathy image more classification methods according to claim 6 based on deep learning, feature
Be: the more disaggregated models of diabetic retinopathy image further include for preventing over-fitting and increasing nonlinear
Dropout layers, described dropout layers is arranged after the full articulamentum of third.
8. diabetic retinopathy image more classification methods according to claim 6 based on deep learning, feature
Be: each maximum pond layer is connected by way of locally connecting and weight is shared with the convolutional layer of its two sides；
It is provided with after each convolutional layer, the first full articulamentum, the second full articulamentum and the full articulamentum of the third
For guaranteeing the ReLU activation primitive of Neural Network Based Nonlinear.
9. diabetic retinopathy image more classification methods according to claim 6 based on deep learning, feature
Be: the output of the full articulamentum of third divides five kinds of classifications of disease using 5 neurons and softmax function
Class.
10. diabetic retinopathy image more classification methods according to claim 6 based on deep learning, feature
Be: the algorithm of the more disaggregated models of diabetic retinopathy image includes:
Training set is made of m marked samples: { (x(1), y(1)) ..., (x(m), y(m)), wherein input feature vector is x(i)∈
Rn+1Classification results are y(i)∈ { 1,2,3,4,5 }, then function hθ(x) form is as follows:
Cost function are as follows:
For the minimization problem of J (θ), it is as follows to obtain gradient formula by derivation using the optimization algorithm of iteration:Iteration requires each time
It is updated as follows:By minimizing J (θ), diabetes view is realized
The more disaggregated models of retinopathy image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910495481.0A CN110210570A (en) | 2019-06-10 | 2019-06-10 | The more classification methods of diabetic retinopathy image based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910495481.0A CN110210570A (en) | 2019-06-10 | 2019-06-10 | The more classification methods of diabetic retinopathy image based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110210570A true CN110210570A (en) | 2019-09-06 |
Family
ID=67791563
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910495481.0A Pending CN110210570A (en) | 2019-06-10 | 2019-06-10 | The more classification methods of diabetic retinopathy image based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110210570A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648344A (en) * | 2019-09-12 | 2020-01-03 | 电子科技大学 | Diabetic retinopathy classification device based on local focus characteristics |
| CN110930391A (en) * | 2019-11-26 | 2020-03-27 | 北京华医共享医疗科技有限公司 | Method, device and equipment for realizing medical image auxiliary diagnosis based on VggNet network model and storage medium |
| CN111260640A (en) * | 2020-01-13 | 2020-06-09 | 重庆大学 | Tree generator network gear pitting image measuring method and device based on cyclean |
| CN111753114A (en) * | 2020-06-04 | 2020-10-09 | 北京字节跳动网络技术有限公司 | Image pre-labeling method and device and electronic equipment |
| CN112819797A (en) * | 2021-02-06 | 2021-05-18 | 国药集团基因科技有限公司 | Diabetic retinopathy analysis method, device, system and storage medium |
| CN112861734A (en) * | 2021-02-10 | 2021-05-28 | 北京农业信息技术研究中心 | Trough food residue monitoring method and system |
| CN112950550A (en) * | 2021-02-04 | 2021-06-11 | 广州中医药大学第一附属医院 | Deep learning-based type 2 diabetic nephropathy image classification method |
| CN113160119A (en) * | 2021-02-04 | 2021-07-23 | 广州中医药大学第一附属医院 | Diabetic retinopathy image classification method based on deep learning |
| CN113538349A (en) * | 2021-06-29 | 2021-10-22 | 山东师范大学 | Small sample diabetic retinopathy classification system based on model independent learning |
| CN113723295A (en) * | 2021-08-31 | 2021-11-30 | 浙江大学 | Face counterfeiting detection method based on image domain frequency domain double-flow network |
| CN113723451A (en) * | 2021-07-20 | 2021-11-30 | 山东师范大学 | Retinal image classification model training method, system, storage medium and device |
| CN114898172A (en) * | 2022-04-08 | 2022-08-12 | 辽宁师范大学 | Diabetic retinopathy classification modeling method based on multi-feature DAG network |
| CN117496231A (en) * | 2023-10-24 | 2024-02-02 | 国药(武汉)精准医疗科技有限公司 | Image classification method and device for diabetic retinopathy and computer equipment |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787046A (en) * | 2016-02-28 | 2016-07-20 | 华东理工大学 | Imbalanced data sorting system based on unilateral dynamic downsampling |
| CN107657602A (en) * | 2017-08-09 | 2018-02-02 | 武汉科技大学 | Based on the breast structure disorder recognition methods for migrating convolutional neural networks twice |
| CN108009592A (en) * | 2017-12-15 | 2018-05-08 | 云南大学 | A kind of diabetic retinal classification of images method |
| CN108021916A (en) * | 2017-12-31 | 2018-05-11 | 南京航空航天大学 | Deep learning diabetic retinopathy sorting technique based on notice mechanism |
| CN108960257A (en) * | 2018-07-06 | 2018-12-07 | 东北大学 | A kind of diabetic retinopathy grade stage division based on deep learning |
| CN109086799A (en) * | 2018-07-04 | 2018-12-25 | 江苏大学 | A kind of crop leaf disease recognition method based on improvement convolutional neural networks model AlexNet |
| CN109635835A (en) * | 2018-11-08 | 2019-04-16 | 深圳蓝韵医学影像有限公司 | A kind of breast lesion method for detecting area based on deep learning and transfer learning |
| CN109691979A (en) * | 2019-01-07 | 2019-04-30 | 哈尔滨理工大学 | A kind of diabetic retina image lesion classification method based on deep learning |
| CN109859172A (en) * | 2019-01-08 | 2019-06-07 | 浙江大学 | Based on the sugared net lesion of eyeground contrastographic picture deep learning without perfusion area recognition methods |
-
2019
- 2019-06-10 CN CN201910495481.0A patent/CN110210570A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787046A (en) * | 2016-02-28 | 2016-07-20 | 华东理工大学 | Imbalanced data sorting system based on unilateral dynamic downsampling |
| CN107657602A (en) * | 2017-08-09 | 2018-02-02 | 武汉科技大学 | Based on the breast structure disorder recognition methods for migrating convolutional neural networks twice |
| CN108009592A (en) * | 2017-12-15 | 2018-05-08 | 云南大学 | A kind of diabetic retinal classification of images method |
| CN108021916A (en) * | 2017-12-31 | 2018-05-11 | 南京航空航天大学 | Deep learning diabetic retinopathy sorting technique based on notice mechanism |
| CN109086799A (en) * | 2018-07-04 | 2018-12-25 | 江苏大学 | A kind of crop leaf disease recognition method based on improvement convolutional neural networks model AlexNet |
| CN108960257A (en) * | 2018-07-06 | 2018-12-07 | 东北大学 | A kind of diabetic retinopathy grade stage division based on deep learning |
| CN109635835A (en) * | 2018-11-08 | 2019-04-16 | 深圳蓝韵医学影像有限公司 | A kind of breast lesion method for detecting area based on deep learning and transfer learning |
| CN109691979A (en) * | 2019-01-07 | 2019-04-30 | 哈尔滨理工大学 | A kind of diabetic retina image lesion classification method based on deep learning |
| CN109859172A (en) * | 2019-01-08 | 2019-06-07 | 浙江大学 | Based on the sugared net lesion of eyeground contrastographic picture deep learning without perfusion area recognition methods |
Non-Patent Citations (4)
| Title |
|---|
| AFONSO MENEGOLA 等: "Knowledge Transfer for Melanoma Screening with Deep Learning", 《IEEE》 * |
| WEIXIN_43216017: "类别不平衡问题的解决方法：过采样、欠采样和再平衡", 《HTTPS://BLOG.CSDN.NET/WEIXIN_43216017/ARTICLE/DETAILS/86713157》 * |
| 伤心的小屁孩: "降采样和上采样（图像）", 《HTTPS://BLOG.CSDN.NET/U013146742/ARTICLE/DETAILS/51865522?WINZOOM=1》 * |
| 天泽28: "softmax回归(Softmax Regression)", 《HTTPS://BLOG.CSDN.NET/U012328159/ARTICLE/DETAILS/72155874》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648344B (en) * | 2019-09-12 | 2023-01-17 | 电子科技大学 | Diabetes retinopathy classification device based on local focus characteristics |
| CN110648344A (en) * | 2019-09-12 | 2020-01-03 | 电子科技大学 | Diabetic retinopathy classification device based on local focus characteristics |
| CN110930391A (en) * | 2019-11-26 | 2020-03-27 | 北京华医共享医疗科技有限公司 | Method, device and equipment for realizing medical image auxiliary diagnosis based on VggNet network model and storage medium |
| CN111260640A (en) * | 2020-01-13 | 2020-06-09 | 重庆大学 | Tree generator network gear pitting image measuring method and device based on cyclean |
| CN111753114A (en) * | 2020-06-04 | 2020-10-09 | 北京字节跳动网络技术有限公司 | Image pre-labeling method and device and electronic equipment |
| CN112950550B (en) * | 2021-02-04 | 2023-11-14 | 广州中医药大学第一附属医院 | Deep learning-based type 2 diabetes kidney disease image classification method |
| CN112950550A (en) * | 2021-02-04 | 2021-06-11 | 广州中医药大学第一附属医院 | Deep learning-based type 2 diabetic nephropathy image classification method |
| CN113160119A (en) * | 2021-02-04 | 2021-07-23 | 广州中医药大学第一附属医院 | Diabetic retinopathy image classification method based on deep learning |
| CN112819797A (en) * | 2021-02-06 | 2021-05-18 | 国药集团基因科技有限公司 | Diabetic retinopathy analysis method, device, system and storage medium |
| CN112819797B (en) * | 2021-02-06 | 2023-09-19 | 国药集团基因科技有限公司 | Method, device, system and storage medium for analyzing diabetic retinopathy |
| CN112861734A (en) * | 2021-02-10 | 2021-05-28 | 北京农业信息技术研究中心 | Trough food residue monitoring method and system |
| CN113538349A (en) * | 2021-06-29 | 2021-10-22 | 山东师范大学 | Small sample diabetic retinopathy classification system based on model independent learning |
| CN113723451A (en) * | 2021-07-20 | 2021-11-30 | 山东师范大学 | Retinal image classification model training method, system, storage medium and device |
| CN113723295A (en) * | 2021-08-31 | 2021-11-30 | 浙江大学 | Face counterfeiting detection method based on image domain frequency domain double-flow network |
| CN113723295B (en) * | 2021-08-31 | 2023-11-07 | 浙江大学 | Face counterfeiting detection method based on image domain frequency domain double-flow network |
| CN114898172A (en) * | 2022-04-08 | 2022-08-12 | 辽宁师范大学 | Diabetic retinopathy classification modeling method based on multi-feature DAG network |
| CN114898172B (en) * | 2022-04-08 | 2024-04-02 | 辽宁师范大学 | Multi-feature DAG network-based diabetic retinopathy classification modeling method |
| CN117496231A (en) * | 2023-10-24 | 2024-02-02 | 国药(武汉)精准医疗科技有限公司 | Image classification method and device for diabetic retinopathy and computer equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110210570A (en) | The more classification methods of diabetic retinopathy image based on deep learning | |
| CN108021916B (en) | Deep learning diabetic retinopathy sorting technique based on attention mechanism | |
| CN109345538B (en) | Retinal vessel segmentation method based on convolutional neural network | |
| Militante et al. | Plant leaf detection and disease recognition using deep learning | |
| CN108648191B (en) | Pest image recognition method based on Bayesian width residual error neural network | |
| CN111476283A (en) | Glaucoma fundus image identification method based on transfer learning | |
| CN107679525A (en) | Image classification method, device and computer-readable recording medium | |
| CN112132817A (en) | Retina blood vessel segmentation method for fundus image based on mixed attention mechanism | |
| CN108960257A (en) | A kind of diabetic retinopathy grade stage division based on deep learning | |
| CN110309811A (en) | A kind of hyperspectral image classification method based on capsule network | |
| CN106919920A (en) | Scene recognition method based on convolution feature and spatial vision bag of words | |
| CN109033994A (en) | A kind of facial expression recognizing method based on convolutional neural networks | |
| CN109447962A (en) | A kind of eye fundus image hard exudate lesion detection method based on convolutional neural networks | |
| Jayakumari et al. | Automated diabetic retinopathy detection and classification using imagenet convolution neural network using fundus images | |
| CN107704886A (en) | A kind of medical image hierarchy system and method based on depth convolutional neural networks | |
| CN113887410A (en) | Deep learning-based multi-category food material identification system and method | |
| CN113011340B (en) | Cardiovascular operation index risk classification method and system based on retina image | |
| CN110096976A (en) | Human behavior micro-Doppler classification method based on sparse migration network | |
| Mardiana et al. | Herbal Leaves Classification Based on Leaf Image Using CNN Architecture Model VGG16 | |
| Sallam et al. | Diabetic retinopathy grading using resnet convolutional neural network | |
| CN114565620A (en) | Fundus image blood vessel segmentation method based on skeleton prior and contrast loss | |
| Dong et al. | Supervised learning-based retinal vascular segmentation by m-unet full convolutional neural network | |
| Chowdhury et al. | Transfer learning approach for diabetic retinopathy detection using efficient network with 2 phase training | |
| CN116236209A (en) | Method for recognizing motor imagery electroencephalogram characteristics of dynamics change under single-side upper limb motion state | |
| CN115587296A (en) | Automatic crop disease identification method based on migration learning step-by-step identification |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190906 |