CN108510473A - The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth - Google Patents
The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth Download PDFInfo
- Publication number
- CN108510473A CN108510473A CN201810192757.3A CN201810192757A CN108510473A CN 108510473 A CN108510473 A CN 108510473A CN 201810192757 A CN201810192757 A CN 201810192757A CN 108510473 A CN108510473 A CN 108510473A
- Authority
- CN
- China
- Prior art keywords
- convolution
- channel
- fcn
- blood vessel
- image
- 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
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
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
-
- 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
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of FCN retinal images blood vessel segmentation methods that combination depth separates convolution and channel weighting, including：1) CLAHE and Gamma corrections are carried out to enhance contrast to the green channel of eye fundus image；2) in order to adapt to network training, piecemeal is carried out with expanding data to enhanced image；3) by depth separate convolution instead of standard convolution in a manner of to increase network-wide, while introduction passage weighting block, the explicitly dependence in Modelling feature channel in a manner of study improves the resolvability of feature.The two is conjointly employed in FCN networks, is tested in DRIVE databases using expert's Manual Logos result as supervision.The result shows that more accurate retinal images blood vessel segmentation may be implemented in this method, and there is higher robustness.
Description
Technical field
The present invention relates to the FCN retinal images blood vessel segmentation sides that a kind of combination depth separates convolution and channel weighting
Method, it is more excellent than the prior art in terms of sensitivity, specificity and accuracy, there is good segmentation performance, belong to medicine figure
As processing, deep learning field.
Background technology
Studies have shown that the diseases such as diabetic retinopathy, artery sclerosis, leukaemia all can generate shadow to optical fundus blood vessel
It rings, leads to its length, width, the variation of angle and blood vessel hyperplasia.Clinically frequently by eye ground image to disease
Carry out screening, analysis and diagnosis.Therefore, in order to carry out quantitative analysis to disease, optical fundus blood vessel is partitioned into retina correlation work
Committed step in work has directive significance to the diagnosis of human diseases, is the embodiment that science promotes the well-being of mankind.
Eye fundus image blood vessel segmentation problem is by extensive concern, and presently, there are some difficult points：(1) for there are tissue damages
Pathological picture, lesion region therein plays prodigious interference effect to the segmentation of blood vessel；(2) in some eyeground pictures,
Highlighted state, similar reflective phenomenon of taking pictures is presented in the center line of blood vessel, and this phenomenon also results in blood vessel segmentation certain difficulty
Degree；(3) fine vascular and the low contrast of background bring difficulty to the identification of capillary in eye fundus image.
There are many kinds of traditional eye fundus image blood vessel segmentation methods, but there are many drawbacks.Segmentation side manually
Method relies on the technical experience of operator, larger by subjective factor, and repeatability is low, and efficiency is relatively low；Unsupervised dividing method is not
Priori signature information is needed, but it is poor for the pathological picture segmentation effect there are tissue damage；The method for having supervision is main
Feature training grader based on extraction trains grader to achieve the purpose that identify blood vessel and non-vascular for feature
It is required that it is very high, and a large amount of retinal vascular images divided in advance is needed to ensure the accurate of model as training sample
Degree requires medical image relatively high.
In recent years, deep learning algorithm achieves important breakthrough, and abstract further feature is formed by combining shallow-layer feature,
And the distributed nature of data is found accordingly.The it is proposed of this structures of FCN is so that the classification of image level further extends into picture
The classification of plain rank, and compared with using image block classification method, improve processing speed.
Invention content
The present invention proposes a kind of FCN retinal images blood vessel segmentation side of combination depth separable convolution and channel weighting
Method, by depth separate convolution instead of standard convolution in a manner of, while in view of the relation of interdependence between feature channel, drawing
Enter channel weighting module, be embedded into FCN network structures and be trained, obtains the network mould with better feature resolution
Type.
Technical scheme of the present invention includes the following steps：
Step 1：Using the libraries DRIVE eye ground image as experimental subjects, to the green channel of image using CLAHE and
Gamma correction enhancing contrasts；
Step 2：Piecemeal is carried out to enhanced eye ground image in step 1 and realizes data extending；
Step 3：On the basis of U-net network portion structures, by depth separate convolution instead of standard convolution in a manner of,
Increase the width of network.Meanwhile a channel weighting module is introduced in each great-jump-forward junction, it is obtained automatically in a manner of study
The significance level in each feature channel is taken, and is adaptively adjusted the characteristic response in each channel according to significance level, completes channel
The re-calibration of feature.
Step 4：Test set image is tested with trained parted pattern, realizes image medium vessels and background two
Point.
Compared with prior art, the beneficial effects of the invention are as follows：
The present invention carries out the study for having supervision on the basis of low volume data, avoids complicated image processing process.
It is more excellent compared with prior art in sensitivity, specificity and accuracy, and AUC has reached 0.98, illustrates that the present invention has very well
Segmentation performance.
In addition, for some pathological images, it also can preferably be partitioned into minute blood vessel；Due to blood vessel caused by background difference
Discontinuously, a part of blood vessel can be also partitioned into.Relative to expert's label as a result, specificity and accuracy are relatively high, therefore this
Invention also has preferable segmentation result for pathology figure.
Description of the drawings
Fig. 1 overall framework schematic diagrames, i.e. Figure of abstract；
Fig. 2 original image pre-processed results figures；
Fig. 2 (a) original images；
Fig. 2 (b) green channel images；
Fig. 2 (c) CLAHE results；
Fig. 2 (d) Gamma correct result；
Image patches and GroundTruth image patches after Fig. 3 enhancings, the first row are image patch after enhancing,
Second row is corresponding GroundTruth images patch；
Fig. 4 Depthwise separable convolution schematic diagrames；
Fig. 5 SE function structure charts；
Fig. 6 combination depth separates the FCN network structures of convolution and channel weighting.
Specific implementation mode
The present invention is described in further detail With reference to embodiment.
The overall framework schematic diagram of the present invention is as shown in Figure 1, first pre-process to increase the parts of images in the libraries DRIVE
Strong contrast；Then data extending is carried out to adapt to the data scale of network training to pretreated image；Then, with depth
Separable convolution replaces the convolution mode of standard, while in view of the degree that interdepends between feature channel, and introduction passage adds
Module is weighed, is embedded into FCN network structures and is trained, material is thus formed improved FCN networks, and then are trained simultaneously
Obtain network model；Finally, using expert's Manual Logos result as the segmentation performance of goldstandard test network model.
Below in conjunction with the accompanying drawings, the specific implementation process of technical solution of the present invention is illustrated.
1. experimental subjects
The eye fundus image data set of the present invention includes the libraries public database DRIVE and retinal structureization analysis library
(Structured Analysis of the Retina, STARE).The libraries DRIVE include 40 colored eye fundus images, training set
With each 20 images of test set, picture size is 565 × 584 (unit is pixel), and every image is equipped with display effective coverage
Bianry image mask.Further include correspondingly the optical fundus blood vessel binary map of the first expert mark in training set, test set includes
The optical fundus blood vessel binary map of first expert and the second expert mark.Usually using one group of mark figure as goldstandard, another set is used
It is compared in other methods.The expert marked manually is instructed and is trained by veteran oculist
's.There are 20 width images to contain corresponding first expert (Adam Hoover) and the second expert (Valentina in the libraries STARE
Kouznetsova) annotation results, picture size are 700 × 605.
2. image preprocessing
2.1 image enhancement
Algorithm for image enhancement is intended to promote picture quality, and image is made to be more clear in terms of content.In order to accelerate network training
Speed, the present invention choose the higher green channel of DRIVE library eye ground picture contrasts and handle.Due to uneven illumination
The movement of eye during weighing and taking pictures, causes picture quality bad, in order to make network restrain as early as possible, therefore to green channel figure
As having carried out normalization operation.Normalized image is handled using CLAHE later, realizes adaptive histogram equalization
With contrast amplitude limit, the contrast and clarity of retinal images blood vessel are improved.The schools Gamma finally are carried out to the result of CLAHE
Just, the dynamic range of image is improved, realizes contrast stretching, enhances picture contrast.Original image pre-processed results such as Fig. 2 institutes
Show.
2.2 data extending
Data scale is very big on the performance influence of training network, and since retinal images are fewer, mark work takes consumption
Power, and retinal images blood vessel width is changed by a pixel to more than ten of pixel, therefore, network training proposed by the present invention
It is based on the image block i.e. processing of patch.Using 20 width image of training set in the libraries DRIVE as sample, to enhanced each image
The image patch of middle extraction 48 × 48, each image extract 10000 width image patch.Correspondingly, GroundTruth images
Carry out same patch extraction operations.Image patch is as shown in Figure 3.
3 combine the FCN of depth separable convolution and channel weighting
3.1 depth separate convolution
The main function of convolutional layer is feature extraction, and three-dimensional convolution kernel is acted on one group of characteristic pattern by the convolution of standard
On, it needs while the correlation of the correlation on studying space and interchannel.Depth separates convolution (Depthwise
Separable convolution) while executing spatial convoluted, it keeps detaching between channel, then according to depth direction
Carry out convolution.As shown in figure 4, depth convolution (Depthwise convolution) is first carried out, that is, each channel inputted is only
It is vertical to execute spatial convoluted, 1 × 1 convolution (Pointwise convolution) is then executed, the channel output of depth convolution is reflected
It is mapped to new channel space.
Then the correlation of Standard convolution while studying space information and interchannel carries out nonlinear activation to output；It is deep
Degree separate convolution carries out depth convolution first, increase the width of network so that feature extraction more enrich, then directly into
1 × 1 convolution of row carries out nonlinear activation to 1 × 1 convolution results.In ginseng quantitative aspects, it is assumed that there are one the convolution of 3 × 3 sizes
Core, input channel 4, output channel 2.Specifically, the convolution kernel of 23 × 3 sizes traverses every number in 4 channels
According to generate 4 × 2=8 characteristic pattern.And then it merges to obtain 1 again by being superimposed the corresponding characteristic pattern of each input channel
Characteristic pattern.2 required output channels finally can be obtained.And depth is applied to separate convolution, with the convolution kernel of 13 × 3 size
The data for traversing 4 channels have obtained 4 characteristic patterns.Before mixing operation, then traversed with the convolution kernel of 21 × 1 sizes
This 4 characteristic patterns, carry out addition fusion.This process has used 4 × 3 × 3+4 × 2 × 1 × 1=40 parameter, is less than standard
4 × 2 × 3 × 3=48 parameter of convolution.When number of channels is more, the parameter amount of reduction can bigger.Therefore, depth can divide
It is not only able to expand network-wide from convolution, and reduces parameter amount to a certain extent.3.2 channel weighting structures
SE (Squeeze-and-Excitation) module is using the global information of characteristic pattern after convolutional layer dynamically to each
The dependence in channel carries out Nonlinear Modeling, promotes the feature learning ability of network.The module is established on the basis of characteristic dimension
On, the explicitly relation of interdependence between Modelling feature channel.It is logical to get each feature automatically by way of study
The significance level in road allows network selectively to enhance useful feature, useful feature is made to obtain more fully according to significance level
It utilizes.SE modules implement：(1) using global average pond (global average pooling) to convolution
Each channel characteristics of layer output are compressed, and the global receptive field information in input feature vector channel is obtained；(2) it uses and contains
The threshold mechanism of sigmoid activation primitives is that each feature channel generates the weight that can learn, explicitly to each feature channel
Between correlation modeled.For limited model complexity and enhance generalization ability, is connected entirely using two in threshold mechanism
Layer is connect, first full articulamentum (fully connected layer, FC) dimensionality reduction is 1/16, passes through an amendment linear unit
After (Rectified Linear Units, ReLU), second FC restores its dimension.This mode both introduced it is non-linear,
It can be preferably fitted complicated correlation between dimension, while can also reduce parameter amount and calculation amount；(3) sigmoid functions are defeated
The weight gone out is the measurement to each feature channel significance level, and weight is special to previous channel by channel weighting by multiplication
In sign, complete on channel dimension to the re-calibration of primitive character.SE modular structures are as shown in Figure 5.
3.3 combine the FCN of depth separable convolution and channel weighting
Depth is separated convolution and is combined with SE modules by the present invention, and convolution is separated for depth, and depth volume is first carried out
Product increases network-wide, then carries out 1 × 1 convolution, merges channel information；Using SE modules, channel is adjusted in a manner of study
Dependence, realize the recalibration of channel characteristics.
Network structure is as shown in fig. 6, the left-half of network includes two groups of Separable convolutional layers, Depthwise convolution
Core size is 3 × 3, using ReLU functions into line activating after every layer of Separable convolution, after every group of Separable convolution
Connect 2 × 2 maximum value ponds (max pooling) that step-length is 2.In order to reduce the loss of characteristic information, every time it is down-sampled after
All number of channels is doubled.The right half part of network includes two groups of deconvolution (Transposed) layers, and core size is 3 × 3.Due to
The feature of network shallow-layer extraction includes many details, and the feature of deep layer is more abstract, is added to shallow-layer feature channel using SE
Power re-scales channel characteristics in a manner of study, enhances useful feature, inhibits unessential ingredient, then adopted by again
The result of sample and the Shallow High Resolution information after channel weighting are attached (concatenate).It is carried out again after every group of connection
The quantity of Separable convolution twice, Separable convolutional channels has carried out halving processing compared to the number of channels of up-sampling,
Depthwise convolution kernel sizes are 3 × 3, use ReLU functions into line activating equally after every layer of Separable convolution.Most
32 characteristic patterns (feature map) are mapped as 2 by later layer, the Standard convolution (conv) for the use of convolution kernel size being 1 × 1
Characteristic pattern (feature map) realizes that eye fundus image blood vessel and background two are classified.
The combination depth of invention, which separates convolution and the FCN network retinal images blood vessel segmentation methods of channel weighting, to be had
Better feature resolution, segmentation performance are preferable.
The foregoing is only a preferred embodiment of the present invention, is not intended to limit the scope of the present invention, should
Understand, the present invention is not limited to implementation as described herein, the purpose of these implementations description is to help this field
In technical staff put into practice the present invention.Any those of skill in the art are easy to do not departing from spirit and scope of the invention
In the case of be further improved and perfect, therefore the present invention is only by the content of the claims in the present invention and limiting for range
System, intention, which covers, all to be included the alternative in the spirit and scope of the invention being defined by the appended claims and waits
Same scheme.
Claims (6)
1. a kind of combination depth separates the FCN retinal images blood vessel segmentation methods of convolution and channel weighting, including following step
Suddenly：
Step 1：CLAHE and Gamma corrections are carried out to the green channel of the part eye ground image in the libraries DRIVE to enhance pair
Degree of ratio；
Step 2：Data extending is carried out to adapt to the data scale of network training to pretreated image in step 1, and is built
Eye ground image data set；
Step 3：By depth separate convolution instead of standard convolution in a manner of, while in view of interdepending between feature channel
Relationship, introduction passage weighting block are embedded into FCN network structures and form improved FCN networks；
Step 4：The eye ground image data set obtained in step 2 is inputted into improved FCN networks and carries out Training,
It is tested in test set using trained parted pattern.
2. combination depth according to claim 1 separates the FCN retinal images blood vessel segmentations of convolution and channel weighting
Method, which is characterized in that in step 1, choose at the higher green channel of DRIVE library eye ground picture contrasts
Reason, and operation is normalized to green channel images, CLAHE and Gamma corrections are carried out to it later to enhance contrast.
3. combination depth according to claim 1 separates the FCN retinal images blood vessel segmentations of convolution and channel weighting
Method, which is characterized in that in step 2, using 20 width original image of training set in the libraries DRIVE as sample, to enhanced every width figure
The image patch of extraction 48 × 48 as in, each image extract 10000 width image patch.Correspondingly, with expert's hand labeled
Sample carries out same patch extraction operations as GroundTruth images.
4. combination depth according to claim 1 separates the FCN retinal images blood vessel segmentations of convolution and channel weighting
Method, which is characterized in that in step 3, depth convolution is first carried out, increases network-wide, then carries out 1 × 1 convolution, fusion is logical
Road information.
5. combination depth according to claim 1 separates the FCN retinal images blood vessel segmentations of convolution and channel weighting
Method, which is characterized in that in step 3, using SE modules, the dependence in channel is adjusted in a manner of study, realizes that channel is special
The recalibration of sign.
6. combination depth according to claim 1 separates the FCN retinal images blood vessel segmentations of convolution and channel weighting
Method, which is characterized in that in step 4, test is carried out to the network model trained and is carried out pair with expert's Manual Logos result
Than to judge the segmentation performance of this method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810192757.3A CN108510473A (en) | 2018-03-09 | 2018-03-09 | The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810192757.3A CN108510473A (en) | 2018-03-09 | 2018-03-09 | The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108510473A true CN108510473A (en) | 2018-09-07 |
Family
ID=63376351
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810192757.3A Pending CN108510473A (en) | 2018-03-09 | 2018-03-09 | The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108510473A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109472315A (en) * | 2018-11-15 | 2019-03-15 | 江苏木盟智能科技有限公司 | A kind of object detection method and system separating convolution based on depth |
| CN109635862A (en) * | 2018-12-05 | 2019-04-16 | 合肥奥比斯科技有限公司 | Retinopathy of prematurity plus lesion classification method |
| CN109636802A (en) * | 2019-01-18 | 2019-04-16 | 天津工业大学 | Pulmonary parenchyma based on depth convolutional neural networks is through CT image partition method |
| CN109727259A (en) * | 2019-01-07 | 2019-05-07 | 哈尔滨理工大学 | A kind of retinal images partitioning algorithm based on residual error U-NET network |
| CN110097554A (en) * | 2019-04-16 | 2019-08-06 | 东南大学 | The Segmentation Method of Retinal Blood Vessels of convolution is separated based on intensive convolution sum depth |
| CN110163852A (en) * | 2019-05-13 | 2019-08-23 | 北京科技大学 | The real-time sideslip detection method of conveyer belt based on lightweight convolutional neural networks |
| CN110309880A (en) * | 2019-07-01 | 2019-10-08 | 天津工业大学 | A kind of 5 days and 9 days hatching egg embryo's image classification methods based on attention mechanism CNN |
| CN110555831A (en) * | 2019-08-29 | 2019-12-10 | 天津大学 | Drainage pipeline defect segmentation method based on deep learning |
| CN110580445A (en) * | 2019-07-12 | 2019-12-17 | 西北工业大学 | Face key point detection method based on GIoU and weighted NMS improvement |
| CN110706232A (en) * | 2019-09-29 | 2020-01-17 | 五邑大学 | Texture image segmentation method, electronic device and computer storage medium |
| CN111368850A (en) * | 2018-12-25 | 2020-07-03 | 展讯通信(天津)有限公司 | Image feature extraction method, image target detection method, image feature extraction device, image target detection device, convolution device, CNN network device and terminal |
| GB2580671A (en) * | 2019-01-22 | 2020-07-29 | Toshiba Kk | A computer vision system and method |
| CN111598894A (en) * | 2020-04-17 | 2020-08-28 | 哈尔滨工业大学 | Retina blood vessel image segmentation system based on global information convolution neural network |
| CN111738949A (en) * | 2020-06-19 | 2020-10-02 | 北京百度网讯科技有限公司 | Image brightness adjusting method and device, electronic equipment and storage medium |
| CN111754514A (en) * | 2019-03-28 | 2020-10-09 | 天津工业大学 | Fundus image hemorrhage point segmentation method based on channel weighting and scaled convolution neural network |
| CN111753820A (en) * | 2019-03-27 | 2020-10-09 | 天津工业大学 | Color fundus image cup segmentation method based on deep learning |
| CN111862056A (en) * | 2020-07-23 | 2020-10-30 | 东莞理工学院 | Retinal vessel image segmentation method based on deep learning |
| CN111882566A (en) * | 2020-07-31 | 2020-11-03 | 华南理工大学 | Blood vessel segmentation method, device, equipment and storage medium of retina image |
| WO2021164299A1 (en) * | 2020-02-20 | 2021-08-26 | 苏州浪潮智能科技有限公司 | Image segmentation method, system, and device, and readable storage medium |
| CN113393446A (en) * | 2021-06-21 | 2021-09-14 | 湖南大学 | Convolutional neural network medical image key point detection method based on attention mechanism |
| CN113593672A (en) * | 2021-04-23 | 2021-11-02 | 安徽医科大学第一附属医院 | Intelligent delineation method for radiotherapy target area |
| US11854247B2 (en) | 2019-04-26 | 2023-12-26 | Tencent Technology (Shenzhen) Company Limited | Data processing method and device for generating face image and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127709A (en) * | 2016-06-24 | 2016-11-16 | 华东师范大学 | A kind of low-luminance color eye fundus image determination methods and Enhancement Method |
| 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 |
| CN106920227A (en) * | 2016-12-27 | 2017-07-04 | 北京工业大学 | Based on the Segmentation Method of Retinal Blood Vessels that deep learning is combined with conventional method |
-
2018
- 2018-03-09 CN CN201810192757.3A patent/CN108510473A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127709A (en) * | 2016-06-24 | 2016-11-16 | 华东师范大学 | A kind of low-luminance color eye fundus image determination methods and Enhancement Method |
| 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 |
| CN106920227A (en) * | 2016-12-27 | 2017-07-04 | 北京工业大学 | Based on the Segmentation Method of Retinal Blood Vessels that deep learning is combined with conventional method |
Non-Patent Citations (3)
| Title |
|---|
| CHANGJIAN SUN ET AL.: "Liver lesion segmentation in CT images with MK-FCN", 《2017 IEEE 2ND ADVANCED INFORMATION TECHNOLOGY, ELECTRONIC AND AUTOMATION CONTROL CONFERENCE (IAEAC)》 * |
| 胡挺: "面向移动平台的轻量级卷积神经网络架构", 《计算机工程-网络首发》 * |
| 郭树旭: "基于全卷积神经网络的肝脏CT影像分割研究", 《计算机工程与应用》 * |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109472315B (en) * | 2018-11-15 | 2021-09-24 | 江苏木盟智能科技有限公司 | Target detection method and system based on depth separable convolution |
| CN109472315A (en) * | 2018-11-15 | 2019-03-15 | 江苏木盟智能科技有限公司 | A kind of object detection method and system separating convolution based on depth |
| CN109635862A (en) * | 2018-12-05 | 2019-04-16 | 合肥奥比斯科技有限公司 | Retinopathy of prematurity plus lesion classification method |
| CN109635862B (en) * | 2018-12-05 | 2021-08-24 | 合肥奥比斯科技有限公司 | Sorting method for retinopathy of prematurity plus lesion |
| CN111368850A (en) * | 2018-12-25 | 2020-07-03 | 展讯通信(天津)有限公司 | Image feature extraction method, image target detection method, image feature extraction device, image target detection device, convolution device, CNN network device and terminal |
| CN111368850B (en) * | 2018-12-25 | 2022-11-25 | 展讯通信(天津)有限公司 | Image feature extraction method, image target detection method, image feature extraction device, image target detection device, convolution device, CNN network device and terminal |
| CN109727259A (en) * | 2019-01-07 | 2019-05-07 | 哈尔滨理工大学 | A kind of retinal images partitioning algorithm based on residual error U-NET network |
| CN109636802A (en) * | 2019-01-18 | 2019-04-16 | 天津工业大学 | Pulmonary parenchyma based on depth convolutional neural networks is through CT image partition method |
| US11315253B2 (en) | 2019-01-22 | 2022-04-26 | Kabushiki Kaisha Toshiba | Computer vision system and method |
| GB2580671B (en) * | 2019-01-22 | 2022-05-04 | Toshiba Kk | A computer vision system and method |
| GB2580671A (en) * | 2019-01-22 | 2020-07-29 | Toshiba Kk | A computer vision system and method |
| CN111753820A (en) * | 2019-03-27 | 2020-10-09 | 天津工业大学 | Color fundus image cup segmentation method based on deep learning |
| CN111754514A (en) * | 2019-03-28 | 2020-10-09 | 天津工业大学 | Fundus image hemorrhage point segmentation method based on channel weighting and scaled convolution neural network |
| CN110097554A (en) * | 2019-04-16 | 2019-08-06 | 东南大学 | The Segmentation Method of Retinal Blood Vessels of convolution is separated based on intensive convolution sum depth |
| US11854247B2 (en) | 2019-04-26 | 2023-12-26 | Tencent Technology (Shenzhen) Company Limited | Data processing method and device for generating face image and medium |
| CN110163852A (en) * | 2019-05-13 | 2019-08-23 | 北京科技大学 | The real-time sideslip detection method of conveyer belt based on lightweight convolutional neural networks |
| CN110163852B (en) * | 2019-05-13 | 2021-10-15 | 北京科技大学 | Conveying belt real-time deviation detection method based on lightweight convolutional neural network |
| CN110309880B (en) * | 2019-07-01 | 2022-12-09 | 天津工业大学 | Method for classifying images of 5-day and 9-day incubated egg embryos based on attention mechanism CNN |
| CN110309880A (en) * | 2019-07-01 | 2019-10-08 | 天津工业大学 | A kind of 5 days and 9 days hatching egg embryo's image classification methods based on attention mechanism CNN |
| CN110580445B (en) * | 2019-07-12 | 2023-02-07 | 西北工业大学 | Face key point detection method based on GIoU and weighted NMS improvement |
| CN110580445A (en) * | 2019-07-12 | 2019-12-17 | 西北工业大学 | Face key point detection method based on GIoU and weighted NMS improvement |
| CN110555831B (en) * | 2019-08-29 | 2023-09-26 | 天津大学 | Deep learning-based drainage pipeline defect segmentation method |
| CN110555831A (en) * | 2019-08-29 | 2019-12-10 | 天津大学 | Drainage pipeline defect segmentation method based on deep learning |
| CN110706232A (en) * | 2019-09-29 | 2020-01-17 | 五邑大学 | Texture image segmentation method, electronic device and computer storage medium |
| WO2021164299A1 (en) * | 2020-02-20 | 2021-08-26 | 苏州浪潮智能科技有限公司 | Image segmentation method, system, and device, and readable storage medium |
| CN111598894B (en) * | 2020-04-17 | 2021-02-09 | 哈尔滨工业大学 | Retina blood vessel image segmentation system based on global information convolution neural network |
| CN111598894A (en) * | 2020-04-17 | 2020-08-28 | 哈尔滨工业大学 | Retina blood vessel image segmentation system based on global information convolution neural network |
| CN111738949A (en) * | 2020-06-19 | 2020-10-02 | 北京百度网讯科技有限公司 | Image brightness adjusting method and device, electronic equipment and storage medium |
| CN111738949B (en) * | 2020-06-19 | 2024-04-05 | 北京百度网讯科技有限公司 | Image brightness adjusting method and device, electronic equipment and storage medium |
| CN111862056A (en) * | 2020-07-23 | 2020-10-30 | 东莞理工学院 | Retinal vessel image segmentation method based on deep learning |
| CN111882566B (en) * | 2020-07-31 | 2023-09-19 | 华南理工大学 | Blood vessel segmentation method, device, equipment and storage medium for retina image |
| CN111882566A (en) * | 2020-07-31 | 2020-11-03 | 华南理工大学 | Blood vessel segmentation method, device, equipment and storage medium of retina image |
| CN113593672A (en) * | 2021-04-23 | 2021-11-02 | 安徽医科大学第一附属医院 | Intelligent delineation method for radiotherapy target area |
| CN113393446A (en) * | 2021-06-21 | 2021-09-14 | 湖南大学 | Convolutional neural network medical image key point detection method based on attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108510473A (en) | The FCN retinal images blood vessel segmentations of convolution and channel weighting are separated in conjunction with depth | |
| Du et al. | An overview of multi-modal medical image fusion | |
| CN108198184B (en) | Method and system for vessel segmentation in contrast images | |
| CN106920227B (en) | The Segmentation Method of Retinal Blood Vessels combined based on deep learning with conventional method | |
| CN110390650B (en) | OCT image denoising method based on dense connection and generation countermeasure network | |
| Li et al. | DeepRetina: layer segmentation of retina in OCT images using deep learning | |
| CN110120033A (en) | Based on improved U-Net neural network three-dimensional brain tumor image partition method | |
| CN110276356A (en) | Eye fundus image aneurysms recognition methods based on R-CNN | |
| CN108510502A (en) | Melanoma picture tissue segmentation methods based on deep neural network and system | |
| CN108257134A (en) | Nasopharyngeal Carcinoma Lesions automatic division method and system based on deep learning | |
| CN108764342B (en) | Semantic segmentation method for optic discs and optic cups in fundus image | |
| CN106600571A (en) | Brain tumor automatic segmentation method through fusion of full convolutional neural network and conditional random field | |
| CN107657612A (en) | Suitable for full-automatic the retinal vessel analysis method and system of intelligent and portable equipment | |
| CN107180421A (en) | A kind of eye fundus image lesion detection method and device | |
| CN109472781A (en) | A kind of diabetic retinopathy detection system based on serial structure segmentation | |
| Deligiannidis et al. | Emerging trends in image processing, computer vision and pattern recognition | |
| CN110786824B (en) | Coarse marking fundus oculi illumination bleeding lesion detection method and system based on bounding box correction network | |
| CN103679801B (en) | A kind of cardiovascular three-dimensional rebuilding method based on various visual angles X-ray | |
| CN102567734B (en) | Specific value based retina thin blood vessel segmentation method | |
| CN109671049A (en) | A kind of medical image processing method, system, equipment, storage medium | |
| CN112598619A (en) | Method for establishing intracranial vascular simulation three-dimensional narrowing model based on transfer learning | |
| CN110110782A (en) | Retinal fundus images optic disk localization method based on deep learning | |
| Lim et al. | Generative data augmentation for diabetic retinopathy classification | |
| Kumar et al. | Image processing in diabetic related causes | |
| CN112509075A (en) | Intracranial vascular lesion marking and three-dimensional display method based on intelligent medical treatment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180907 |