CN107704886A - A kind of medical image hierarchy system and method based on depth convolutional neural networks - Google Patents
A kind of medical image hierarchy system and method based on depth convolutional neural networks Download PDFInfo
- Publication number
- CN107704886A CN107704886A CN201710982421.2A CN201710982421A CN107704886A CN 107704886 A CN107704886 A CN 107704886A CN 201710982421 A CN201710982421 A CN 201710982421A CN 107704886 A CN107704886 A CN 107704886A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- image
- depth convolutional
- classifier
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims description 25
- 239000000835 fiber Substances 0.000 claims abstract description 19
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 17
- 230000007547 defect Effects 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 16
- 239000000284 extract Substances 0.000 claims abstract description 12
- 230000011218 segmentation Effects 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 3
- 210000003733 optic disk Anatomy 0.000 abstract description 5
- 238000003745 diagnosis Methods 0.000 abstract description 3
- 208000010412 Glaucoma Diseases 0.000 description 23
- 230000006870 function Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000007477 logistic regression Methods 0.000 description 5
- 230000004044 response Effects 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 230000007935 neutral effect Effects 0.000 description 4
- 210000004204 blood vessel Anatomy 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 210000004126 nerve fiber Anatomy 0.000 description 2
- 210000001328 optic nerve Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000002688 persistence Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 210000001364 upper extremity Anatomy 0.000 description 2
- 206010003694 Atrophy Diseases 0.000 description 1
- 210000001367 artery Anatomy 0.000 description 1
- 230000037444 atrophy Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000002715 modification method Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- ORQBXQOJMQIAOY-UHFFFAOYSA-N nobelium Chemical compound [No] ORQBXQOJMQIAOY-UHFFFAOYSA-N 0.000 description 1
- 208000001749 optic atrophy Diseases 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000002207 retinal effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 210000003462 vein Anatomy 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of medical image stage division and system based on depth convolutional neural networks, including：The automatic segmentation of original eye fundus image, it is divided into optic disk optic cup two parts image；Extract image green channel component；The gray level image extracted using histogram equalization amendment；Two kinds of cup disc ratio, layer of optic fibers defect features are extracted respectively；Use the multiple sub-classifiers of depth convolutional neural networks Algorithm for Training；Sub-classifier is combined, ballot draws final classification result.Using technical scheme, classification accuracy is obviously improved, and helps to reduce mistaken diagnosis, so as to improve the practical value of grader.
Description
Technical field
The invention belongs to machine learning field, more particularly to a kind of medical image (pin based on depth convolutional neural networks
To glaucoma eye fundus image) hierarchy system and method
Background technology
Glaucoma is second largest blinding disease in the world.The eyeground of glaucoma patients generally has big optic disk ratio, arteries and veins
Network film atrophy arc and nerve fiber layer defects.Glaucoma screening is an extremely complex and difficult task.Glaucoma at present
Diagnosis be substantially by artificial observation mode.But its weak point be present in the vision system of people, such as in the presence of subjectivity
Property, limitation, ambiguity, lack persistence etc..In order to realize intelligent automaticization of detection and informationization, an urgent demand
The computer picture recognition of a kind of visual performance that can simulate people and the performance for surmounting it is with diagnostic system to identify and examine
Disconnected glaucoma lesion.
The content of the invention
The problem of being identified for glaucoma eye fundus image, the present invention provide a kind of medical treatment based on depth convolutional neural networks
Image grading system and method.
The present invention proposes a kind of eye fundus image recognition methods, has used new preprocess method and new depth deeper
Neutral net data set is handled, pretreatment link add extraction G passages, histogram equalization, improve it is original
The steps such as the quantity-quality of data set, 8 layers of instead preceding 5 layer networks of Alexnet are employed in terms of neutral net.Except this
Outside also added " layer of optic fibers defect " this feature glaucoma judged.Improve the overall accurate of system
Rate.This has very strong practical significance for the practical application of grader.
To achieve the above object, the present invention adopts the following technical scheme that：
The present invention provides a kind of medical image stage division based on depth convolutional neural networks, including：
Automatic segmentation original image
Extract original medical image green channel component；
The gray level image extracted using histogram equalization amendment；
Respectively from revised image zooming-out cup disc ratio feature and layer of optic fibers defect feature；
Character subset is respectively trained using depth convolutional neural networks algorithm and produces multiple sub-classifiers；
Sub-classifier is combined, ballot draws final classification result.
Preferably, the automatic segmentation original image is to automatically extract technology using ROI to realize, according to the spy to be extracted
The characteristics of sign, extracted region interested in image is come out automatically, computer detects the position and face of optic cup optic disk automatically
Volume data, the image manipulation after being carry out element task.
Preferably, the green that the green channel component, which is colored medical image, to be contained in 3 components of red, green, blue is divided
Amount.
Preferably, the histogram equalization turns to the method that picture contrast quality is automatically adjusted using greyscale transformation.
Preferably, the gray level image extracts green channel component image.
Preferably, the cup disc ratio feature, layer of optic fibers defect feature are respectively：Medical image is according to cup disc ratio
The feature that is extracted after analyzing and processing, the feature extracted after being analyzed and processed according to layer of optic fibers defect.
Preferably, it is described using depth convolutional neural networks algorithm be respectively trained character subset produce sub-classifier be
The image pre-processed is inputted into depth convolutional neural networks Algorithm for Training, generates corresponding sub-classifier.
Preferably, the combination sub-classifier, ballot show that final classification result is：Medical image is tested respectively by instructing
The sub-classifier classification perfected, statistical classification result, most multiclass are final classification result.
The present invention also provides a kind of medical image hierarchy system based on depth convolutional neural networks, including：
Automatic segmentation original image device, it is configured with automatically extracting technology extraction region of interest area image；
Green channel classification extraction element, is configured as extracting original medical image green channel component；
Histogram equalization device, it is configured to, with the gray level image that histogram equalization amendment extracts；
Feature deriving means, it is configured to from revised image zooming-out cup disc ratio feature；Optic nerve fiber break
Damage feature；
Sub-classifier trainer, it is configured with depth convolutional neural networks algorithm and character subset generation is respectively trained
Sub-classifier；
As a result balloting device, it is configured as combining sub-classifier, ballot draws final classification result.
Preferably, the automatic segmentation original image is to automatically extract technology using ROI to realize.
Preferably, the green that the green channel component, which is colored medical image, to be contained in 3 components of red, green, blue is divided
Amount.
Preferably, the histogram equalization is a kind of side that picture contrast quality is automatically adjusted using greyscale transformation
Method.
Preferably, the gray level image extracts green channel component image.
Preferably, the cup disc ratio feature, layer of optic fibers defect feature are respectively：Medical image is according to cup disc ratio
The feature that is extracted after analyzing and processing, the feature extracted after being analyzed and processed according to layer of optic fibers defect.
Preferably, it is described using depth convolutional neural networks algorithm be respectively trained character subset produce sub-classifier be
The image pre-processed is inputted into depth convolutional neural networks Algorithm for Training, generates corresponding sub-classifier.
Preferably, the combination sub-classifier, ballot show that final classification result is：Medical image is tested respectively by instructing
The sub-classifier classification perfected, statistical classification result, most multiclass are final classification result.
The present invention proposes a kind of eye fundus image recognition methods and system, has used new preprocess method and new depth
Spend deeper neutral net to handle data set, added in pretreatment link and extract G passages, histogram equalization, carried
The steps such as the quantity-quality of high original data set, 8 layers of the instead preceding 5 layers of nets of Alexnet are employed in terms of neutral net
Network.In addition we also added " layer of optic fibers defect " this feature and glaucoma judged.Improve system
Overall accuracy rate.This has very strong practical significance for the practical application of grader.
Brief description of the drawings
With reference to accompanying drawing, from the following detailed description to the embodiment of the present invention, the present invention is better understood with, it is similar in accompanying drawing
Label indicate similar part, wherein：
Fig. 1 shows the glaucoma eye fundus image point according to an embodiment of the invention based on depth convolutional neural networks
One detailed diagram of level system；
Fig. 2 shows the glaucoma eye fundus image according to an embodiment of the invention based on depth convolutional neural networks
One detailed diagram of stage division；
Fig. 3 shows the schematic diagram of eight layer depths convolutional neural networks according to an embodiment of the invention.
Embodiment
The feature and exemplary embodiment of various aspects of the present invention is described more fully below.Following description covers many
Detail, to provide complete understanding of the present invention.It will be apparent, however, to one skilled in the art that
The present invention can be implemented in the case of some details in not needing these details.Description to embodiment below is only
It is to be provided by showing the example of the present invention to the clearer understanding of the present invention.The present invention is not limited to set forth below
Any concrete configuration and algorithm, but cover coherent element, part and calculation under the premise of without departing from the spirit of the present invention
Any modification, replacement and the improvement of method.
Multiple problems in view of the above, the present invention propose a kind of glaucoma eye based on depth convolutional neural networks
Base map is as hierarchy system.With reference to Fig. 1 and Fig. 2, illustrate according to classification of the present invention based on unbalanced medical image data sets
The example of method and system.Fig. 1 shows the glaucoma according to an embodiment of the invention based on depth convolutional neural networks
One detailed diagram of eye fundus image hierarchy system；Fig. 2 shows according to an embodiment of the invention based on depth convolution god
One detailed diagram of the sorting technique of the glaucoma eye fundus image through network；
As shown in figure 1, image is included certainly according to a kind of categorizing system based on unbalanced medical image data sets of the present invention
Dynamic segmenting device 101, green channel classification extraction element 102, histogram equalization device 103, feature deriving means 104, son
Classifier training device 105, result balloting device 106.Their function is as follows：Automatically split original eye fundus image (that is, to perform
Step S201).Extract original medical image green channel component (that is, performing step S202).Utilize histogram equalization amendment
The gray level image (that is, performing step S203) extracted.Respectively from revised image zooming-out cup disc ratio feature, optic nerve fiber
Break damage feature (that is, performing step S204).Corresponding sub-classifier is trained (that is, to perform step using depth convolutional neural networks
Rapid S205).To combine sub-classifier, ballot draws final classification result (that is, performing step S206).
Specifically, feature deriving means 104 introduce the concept of cup disc ratio feature and layer of optic fibers defect feature, according to
Glaucoma eye fundus image is judged according to both features.Sub-classifier trainer 105 is then to utilize depth convolution god
Through network model model, using the eye fundus image handled well as input, corresponding sub-classifier is trained.Below, provide by according to this
The example of categorizing system of the invention based on unbalanced medical image data sets：
This introduces detailed process exemplified by sentencing eye fundus image.
Image intercepts one kind side that technology refers to Computer Automatic Recognition and is partitioned into area-of-interest (ROI) automatically
Method, the image of optic cup optic disk part and the image of nerve fibre layer segment can be intercepted according to demand.
Colored eye fundus image contains 3 components of red, green, blue.Due to red component brightness highest, blood vessel and background contrasts
It is low, it is not easy target blood and the differentiation of eyeground background；Blue component contrast and brightness are low, and noise jamming is serious；Green
The brightness of component is moderate, and blood vessel and background contrasts are higher, can react colored optical fundus blood vessel distribution very well.So to training
Collection extraction green channel (G passages) component.
Histogram equalization is a kind of method that picture contrast quality is automatically adjusted using greyscale transformation, and basic thought is
Greyscale transformation function is obtained by the probability density function of gray level, it is one kind based on Cumulative Distribution Function transform method
Histogram Modification Methods.So the training set extraction sorted gray level image of green channel is corrected using histogram equalization
Image.
Glaucoma is second largest blinding disease in the world, and screening glaucoma is an extremely complex and difficult task.
The diagnosis of glaucoma is substantially by artificial observation mode at present.But its weak point be present in the vision system of people, than
Subjectivity, limitation, ambiguity, shortage persistence such as be present.
The three of glaucoma are characterized as greatly：
(1) cup disc ratio>=0.7 (in normal eye, the area ratio of optic cup and optic disk is smaller, less than 0.6, is occurring
In the eyes of glaucoma, the missing of nerve fibre causes cup area to increase, so that cup disc ratio increases；Participating in the big of statistics
In the majority person of being observed, CDR occurs in only general 5% non-glaucomatous>=0.7 situation).
(2) along ISNT rules are not met, (most of normal optical papillae all meets ＂ ISNT ＂ rules to disk：Lower edge (INFERIOR)
Thicker than upper limb (SUPERIOR), upper limb is thicker than nasal side (NASAL) side, and top side (TEMPORAL) is most thin).
(3) layer of optic fibers defect, optic atrophy (most of such case is betided in glaucoma).Glaucoma meeting
It is gradually reduced retinal nerve fiber layer thickness, defect.
What we used in the present invention is feature (1), (3), i.e. cup disc ratio feature and layer of optic fibers defect feature.
For revised gray level image, extraction feature set is handled using depth convolutional neural networks, training classification afterwards
Device.Training set now is changed into two independent feature sets, respectively cup disc ratio feature set, layer of optic fibers defect feature
Collection.
The two category feature data sets obtained afterwards by previous step sample the character subset obtained, use depth convolutional Neural net
Network learning algorithm, which is respectively trained, obtains multiple and character subset and separate sub-classifier.
Network training method in glaucoma eye fundus image detection based on convolutional neural networks comprises the following steps：
(1) original image is pre-processed accordingly, including Image Automatic Segmentation, extraction green channel, histogram are equal
Four steps of weighing apparatusization and feature extraction.
(2) one 8 layers of depth convolutional neural networks are constructed, such as figure three.8 Rotating fields of the model in general, are not wrapped
Input layer is included, 1-5 layers are convolutional layers, and 6-8 layers are then full articulamentum, and last output layer is also full articulamentum, and can be seen
Make the softmax graders of 1000 dimension outputs, last optimization aim is to maximize average multinomial logistic regression.In convolutional layer
1st, also include sublayer convolutional layer 1 and convolutional layer 2 in 2 structure, be in response to normalize layer afterwards, be i.e. response normalization layer 1, response
Normalize 2 layers of layer.The operation closelyed follow after each convolutional layer and full articulamentum is ReLU operations.Maximum pond operation is tight
Follow in first response normalization layer 1, response normalization layer 2, and the 5th convolutional layer.Dropout operation be most latter two
Full articulamentum.Pondization operation (Pooling) is used for after convolution operation, and pond layer is not belonging to single in CNN with convolutional layer
Layer, is not charged in the CNN number of plies yet.
Wherein, input and output expression formula is corresponding to convolution node layer：
Input and output expression formula is corresponding to the node layer of pond：
In above formula, l is the network number of plies, and j is j-th of characteristic pattern number of l layers,For the output on j-th of characteristic pattern of l layers
Matrix,L-1 layer ith feature figure matrixes, here as input matrix are represented,It is and l-1 layer ith feature figures
The parameter matrix of j-th of characteristic pattern of corresponding l layers,Weights corresponding to l-th of sample level, j-th of characteristic pattern are represented, f is corresponded to
Sigmoid functions.
Expression formula corresponding to logistic regression layer is：
In above formula, X is the value of logistic regression layer input node, and θ is that logistic regression layer treats training parameter, and h θ (x) are logic
The output of layer is returned, implication is the probability that output node takes 1.Local acknowledgement normalizes, after the position of effect is ReLU, specifically
Formula is：
In above formula,It is i in x, the Feature Mapping of y-coordinate, j x, other adjacent Feature Mappings of y location.
Softmax loss function, calculation formula are：
In above formula, (i, j) is i-th, j element in matrix f (x, w).max(fj) i.e. at all points of i-th of sample
Maximum score in class score.
(3) training image is input in network and convolutional neural networks is trained.Using softmax loss conducts
Loss function, using back-propagation algorithm, parameter in the average loss function training convolutional neural networks in training with
And logistic regression layer parameter produces independent sub-classifier to network convergence.
Most whole mutually independent sub-classifiers are combined at last, test the eye fundus image sub-classifier by training respectively
Classification, statistical classification result, most multiclass are final classification result.
This method and system are applicable not only to eye fundus image classification, and other medical image classification are applicable.
Need clearly, the invention is not limited in particular configuration that is described above and being shown in figure and processing.Also,
For brevity, the detailed description to known method technology is omitted here.In the above-described embodiments, have been described and illustrated some
Specific step is as example.But procedure of the invention is not limited to described and illustrated specific steps, this area
Technical staff can understand the present invention spirit after, be variously modified, change and add, or change step between
Order.
Functional block shown in structures described above block diagram can be implemented as hardware, software, firmware or their group
Close.When realizing in hardware, its may, for example, be electronic circuit, application specific integrated circuit (ASIC), appropriate firmware, insert
Part, function card etc..When being realized with software mode, element of the invention is used to perform program or the generation of required task
Code section.Either code segment can be stored in machine readable media program or the data-signal by being carried in carrier wave is passing
Defeated medium or communication links are sent." machine readable media " can include any medium that can store or transmit information.
The example of machine readable media includes electronic circuit, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), soft
Disk, CD-ROM, CD, hard disk, fiber medium, radio frequency (RF) link, etc..Code segment can be via such as internet, inline
The computer network of net etc. is downloaded.
The present invention can realize in other specific forms, without departing from its spirit and essential characteristics.For example, particular implementation
Algorithm described in example can be changed, and system architecture is without departing from the essence spirit of the present invention.Therefore, currently
Embodiment be all counted as being exemplary rather than in all respects it is limited, the scope of the present invention by appended claims rather than
Foregoing description defines, also, fall into claim implication and equivalent in the range of whole change so as to all be included in
Among the scope of the present invention.
Claims (8)
- A kind of 1. medical image sorting technique based on depth convolutional neural networks, it is characterised in that including：Automatic segmentation original imageExtract original medical image green channel component；The gray level image extracted using histogram equalization amendment；Respectively from revised image zooming-out cup disc ratio feature and layer of optic fibers defect feature；Character subset is respectively trained using depth convolutional neural networks algorithm and produces multiple sub-classifiers；Sub-classifier is combined, ballot draws final classification result.
- 2. the medical image stage division according to claim 1 based on depth convolutional neural networks, it is characterised in that institute Stating automatic segmentation original image process is：Technology is automatically extracted using area-of-interest (ROI) to realize.
- 3. the medical image stage division according to claim 1 based on depth convolutional neural networks, it is characterised in that institute State using depth convolutional neural networks algorithm be respectively trained character subset produce sub-classifier be will the image that pre-process it is defeated Enter depth convolutional neural networks Algorithm for Training, generate corresponding sub-classifier.
- 4. the medical image stage division according to claim 1 based on depth convolutional neural networks, it is characterised in that institute Combination sub-classifier is stated, ballot show that final classification result is：Test medical image is classified by the sub-classifier trained respectively, Statistical classification result, most multiclass are final classification result.
- A kind of 5. medical image hierarchy system based on depth convolutional neural networks, it is characterised in that including：Automatic segmentation original image device, it is configured with automatically extracting technology extraction region of interest area image；Green channel classification extraction element, is configured as extracting original medical image green channel component；Histogram equalization device, it is configured to, with the gray level image that histogram equalization amendment extracts；Feature deriving means, it is configured to from revised image zooming-out cup disc ratio feature；Layer of optic fibers defect is special Sign；Sub-classifier trainer, it is configured with depth convolutional neural networks algorithm and character subset generation son point is respectively trained Class device；As a result balloting device, it is configured as combining sub-classifier, ballot draws final classification result.
- 6. the medical image hierarchy system according to claim 5 based on depth convolutional neural networks, it is characterised in that institute Stating automatic segmentation original image process is：Area-of-interest (ROI) make use of to automatically extract technology realization.
- 7. the medical image hierarchy system according to claim 5 based on depth convolutional neural networks, it is characterised in that institute State using depth convolutional neural networks algorithm be respectively trained character subset produce sub-classifier be will the image that pre-process it is defeated Enter depth convolutional neural networks Algorithm for Training, generate corresponding sub-classifier.
- 8. the medical image hierarchy system according to claim 5 based on depth convolutional neural networks, it is characterised in that institute Combination sub-classifier is stated, ballot show that final classification result is：Test medical image is classified by the sub-classifier trained respectively, Statistical classification result, most multiclass are final classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710982421.2A CN107704886A (en) | 2017-10-20 | 2017-10-20 | A kind of medical image hierarchy system and method based on depth convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710982421.2A CN107704886A (en) | 2017-10-20 | 2017-10-20 | A kind of medical image hierarchy system and method based on depth convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107704886A true CN107704886A (en) | 2018-02-16 |
Family
ID=61181869
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710982421.2A Pending CN107704886A (en) | 2017-10-20 | 2017-10-20 | A kind of medical image hierarchy system and method based on depth convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107704886A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108492285A (en) * | 2018-03-13 | 2018-09-04 | 北京理工大学 | A kind of radial line optic disk atrophy arc dividing method based on grey scale change |
| CN108596247A (en) * | 2018-04-23 | 2018-09-28 | 南方医科大学 | A method of fusion radiation group and depth convolution feature carry out image classification |
| CN109615614A (en) * | 2018-11-26 | 2019-04-12 | 北京工业大学 | The extracting method and electronic equipment of eye fundus image medium vessels based on multi-feature fusion |
| CN110415252A (en) * | 2018-04-26 | 2019-11-05 | 北京连心医疗科技有限公司 | A kind of eye circumference organ segmentation method, equipment and storage medium based on CNN |
| CN110749554A (en) * | 2018-07-23 | 2020-02-04 | 波音公司 | Characterization of the ratio of melted tissue strands in a layer of fibrous material |
| CN111862187A (en) * | 2020-09-21 | 2020-10-30 | 平安科技（深圳）有限公司 | Cup-to-tray ratio determining method, device, equipment and storage medium based on neural network |
| CN111951933A (en) * | 2020-08-07 | 2020-11-17 | 平安科技（深圳）有限公司 | Eyeground color photograph image grading method, device, computer equipment and storage medium |
| US10963757B2 (en) | 2018-12-14 | 2021-03-30 | Industrial Technology Research Institute | Neural network model fusion method and electronic device using the same |
| CN113454638A (en) * | 2018-12-19 | 2021-09-28 | 艾奎菲股份有限公司 | System and method for joint learning of complex visual inspection tasks using computer vision |
| WO2021189848A1 (en) * | 2020-09-22 | 2021-09-30 | 平安科技（深圳）有限公司 | Model training method and apparatus, cup-to-disc ratio determination method and apparatus, and device and storage medium |
| CN113487582A (en) * | 2021-07-16 | 2021-10-08 | 依未科技(北京)有限公司 | Image evaluation method and device, computer-readable storage medium and electronic device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
| CN106651888A (en) * | 2016-09-28 | 2017-05-10 | 天津工业大学 | Color fundus image optic cup segmentation method based on multi-feature fusion |
| CN106682389A (en) * | 2016-11-18 | 2017-05-17 | 武汉大学 | Health management system for monitoring ocular lesions resulting from high blood pressure |
| CN106878677A (en) * | 2017-01-23 | 2017-06-20 | 西安电子科技大学 | Student classroom Grasping level assessment system and method based on multisensor |
| US20170270653A1 (en) * | 2016-03-15 | 2017-09-21 | International Business Machines Corporation | Retinal image quality assessment, error identification and automatic quality correction |
-
2017
- 2017-10-20 CN CN201710982421.2A patent/CN107704886A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170270653A1 (en) * | 2016-03-15 | 2017-09-21 | International Business Machines Corporation | Retinal image quality assessment, error identification and automatic quality correction |
| CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
| CN106651888A (en) * | 2016-09-28 | 2017-05-10 | 天津工业大学 | Color fundus image optic cup segmentation method based on multi-feature fusion |
| CN106682389A (en) * | 2016-11-18 | 2017-05-17 | 武汉大学 | Health management system for monitoring ocular lesions resulting from high blood pressure |
| CN106878677A (en) * | 2017-01-23 | 2017-06-20 | 西安电子科技大学 | Student classroom Grasping level assessment system and method based on multisensor |
Non-Patent Citations (1)
| Title |
|---|
| 姜玉苹等: "基于眼底图像的人工神经网络分类器研究现状与进展", 《中国数字医学》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108492285A (en) * | 2018-03-13 | 2018-09-04 | 北京理工大学 | A kind of radial line optic disk atrophy arc dividing method based on grey scale change |
| CN108596247A (en) * | 2018-04-23 | 2018-09-28 | 南方医科大学 | A method of fusion radiation group and depth convolution feature carry out image classification |
| CN110415252B (en) * | 2018-04-26 | 2022-08-05 | 北京连心医疗科技有限公司 | CNN-based periocular organ segmentation method, CNN-based periocular organ segmentation equipment and CNN-based periocular organ segmentation storage medium |
| CN110415252A (en) * | 2018-04-26 | 2019-11-05 | 北京连心医疗科技有限公司 | A kind of eye circumference organ segmentation method, equipment and storage medium based on CNN |
| CN110749554A (en) * | 2018-07-23 | 2020-02-04 | 波音公司 | Characterization of the ratio of melted tissue strands in a layer of fibrous material |
| CN109615614A (en) * | 2018-11-26 | 2019-04-12 | 北京工业大学 | The extracting method and electronic equipment of eye fundus image medium vessels based on multi-feature fusion |
| CN109615614B (en) * | 2018-11-26 | 2020-08-18 | 北京工业大学 | Method for extracting blood vessels in fundus image based on multi-feature fusion and electronic equipment |
| US10963757B2 (en) | 2018-12-14 | 2021-03-30 | Industrial Technology Research Institute | Neural network model fusion method and electronic device using the same |
| CN113454638A (en) * | 2018-12-19 | 2021-09-28 | 艾奎菲股份有限公司 | System and method for joint learning of complex visual inspection tasks using computer vision |
| CN111951933B (en) * | 2020-08-07 | 2023-01-17 | 平安科技（深圳）有限公司 | Eyeground color photograph image grading method, device, computer equipment and storage medium |
| CN111951933A (en) * | 2020-08-07 | 2020-11-17 | 平安科技（深圳）有限公司 | Eyeground color photograph image grading method, device, computer equipment and storage medium |
| CN111862187B (en) * | 2020-09-21 | 2021-01-01 | 平安科技（深圳）有限公司 | Cup-to-tray ratio determining method, device, equipment and storage medium based on neural network |
| WO2021189849A1 (en) * | 2020-09-21 | 2021-09-30 | 平安科技（深圳）有限公司 | Cup-to-disc ratio determination method and apparatus based on neural network, and device and storage medium |
| JP2022533685A (en) * | 2020-09-21 | 2022-07-25 | 平安科技（深▲せん▼）有限公司 | Neural network-based C/D ratio determination method, apparatus, equipment, and storage medium |
| CN111862187A (en) * | 2020-09-21 | 2020-10-30 | 平安科技（深圳）有限公司 | Cup-to-tray ratio determining method, device, equipment and storage medium based on neural network |
| JP7260674B2 (en) | 2020-09-21 | 2023-04-18 | 平安科技（深▲せん▼）有限公司 | C/D ratio determination method based on neural network, computer equipment and storage medium |
| WO2021189848A1 (en) * | 2020-09-22 | 2021-09-30 | 平安科技（深圳）有限公司 | Model training method and apparatus, cup-to-disc ratio determination method and apparatus, and device and storage medium |
| CN113487582A (en) * | 2021-07-16 | 2021-10-08 | 依未科技(北京)有限公司 | Image evaluation method and device, computer-readable storage medium and electronic device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107704886A (en) | A kind of medical image hierarchy system and method based on depth convolutional neural networks | |
| CN112132817B (en) | Retina blood vessel segmentation method for fundus image based on mixed attention mechanism | |
| WO2021253939A1 (en) | Rough set-based neural network method for segmenting fundus retinal vascular image | |
| CN107330889B (en) | A kind of Chinese medicine tongue color coating colour automatic analysis method based on convolutional neural networks | |
| CN110276356A (en) | Eye fundus image aneurysms recognition methods based on R-CNN | |
| CN109345538A (en) | A kind of Segmentation Method of Retinal Blood Vessels based on convolutional neural networks | |
| CN108198207A (en) | Multiple mobile object tracking based on improved Vibe models and BP neural network | |
| Karegowda et al. | Exudates detection in retinal images using back propagation neural network | |
| Agrawal et al. | Grape leaf disease detection and classification using multi-class support vector machine | |
| CN109635846A (en) | A kind of multiclass medical image judgment method and system | |
| CN108021916A (en) | Deep learning diabetic retinopathy sorting technique based on notice mechanism | |
| CN108095683A (en) | The method and apparatus of processing eye fundus image based on deep learning | |
| CN107330892A (en) | A kind of sunflower disease recognition method based on random forest method | |
| CN106530283A (en) | SVM (support vector machine)-based medical image blood vessel recognition method | |
| CN109711426A (en) | A kind of pathological picture sorter and method based on GAN and transfer learning | |
| CN106529598A (en) | Classification method and system based on imbalanced medical image data set | |
| Chen et al. | Cell nuclei detection and segmentation for computational pathology using deep learning | |
| CN108427963A (en) | A kind of dermopathic classifying identification method of melanoma based on deep learning | |
| CN107748889A (en) | A kind of Ultrasound Image of Breast Tumor automatic classification method | |
| CN110163839A (en) | The recognition methods of leopard line shape eye fundus image, model training method and equipment | |
| Bhatkalkar et al. | Automated fundus image quality assessment and segmentation of optic disc using convolutional neural networks | |
| Erwin et al. | Hybrid multilevel thresholding and improved harmony search algorithm for segmentation | |
| CN110610480A (en) | MCASPP neural network eyeground image optic cup optic disc segmentation model based on Attention mechanism | |
| CN106846301A (en) | Retinal images sorting technique and device | |
| Iqbal et al. | Automatic diagnosis of diabetic retinopathy using fundus images |
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: 20180216 |