CN107369160A - A kind of OCT image median nexus film new vessels partitioning algorithm - Google Patents

A kind of OCT image median nexus film new vessels partitioning algorithm Download PDF

Info

Publication number
CN107369160A
CN107369160A CN201710512064.3A CN201710512064A CN107369160A CN 107369160 A CN107369160 A CN 107369160A CN 201710512064 A CN201710512064 A CN 201710512064A CN 107369160 A CN107369160 A CN 107369160A
Authority
CN
China
Prior art keywords
image
super
pixel
formula
new vessels
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.)
Granted
Application number
CN201710512064.3A
Other languages
Chinese (zh)
Other versions
CN107369160B (en
Inventor
陈新建
袭肖明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Were Medical Technology Co Ltd
Original Assignee
Suzhou Were Medical Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou Were Medical Technology Co Ltd filed Critical Suzhou Were Medical Technology Co Ltd
Priority to CN201710512064.3A priority Critical patent/CN107369160B/en
Publication of CN107369160A publication Critical patent/CN107369160A/en
Priority to PCT/CN2018/089056 priority patent/WO2019001208A1/en
Application granted granted Critical
Publication of CN107369160B publication Critical patent/CN107369160B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • 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
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • 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
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular

Abstract

The invention discloses a kind of OCT image median nexus film new vessels partitioning algorithm, comprise the following steps：S01：A kind of structure priori learning method is designed to training image, builds structure prior matrix, the structure prior matrix is used to distinguish CNV region and background area；S02：OCT original images are converted to by conspicuousness enhancing image based on the structure prior matrix, for strengthening the conspicuousness in CNV region；S03：Multiscale analysis is used on conspicuousness enhancing image, conspicuousness enhancing image is divided into m yardstick；S04：Train to obtain the m convolutional neural networks models trained based on every kind of yardstick；S05：Test image is handled using step S01, S02, S03, tested using the convolutional neural networks model trained in step S04, export m segmentation result, m segmentation result is merged can significantly improve the segmentation precision of OCT image median nexus film new vessels as final segmentation result, the present invention.

Description

A kind of OCT image median nexus film new vessels partitioning algorithm
Technical field
The present invention relates to a kind of OCT image median nexus film new vessels partitioning algorithm, belong to retinal images cutting techniques Field.
Background technology
The existing automatic cutting techniques of CNV are mostly based on Fundus fluorescein image.Compare Compared with Fundus fluorescein, OCT image has the advantages that noninvasive, high speed, high-resolution, three-dimensional imaging, and it is for old age Property the clinical common ophthalmological disorder of denaturation macula lutea etc. auxiliary diagnosis there is prior clinical meaning.
There has been no the CNV partitioning algorithm based on OCT image at present.The new green blood of choroid in OCT image Pipe segmentation faces lot of challenges：Texture variations are larger, inconsistent gray scale inhomogeneity, shapes and sizes be present, obscurity boundary, deposit In a large amount of spot noises etc..These problems make it that traditional method is difficult to obtain more accurately segmentation effect.Convolutional neural networks With powerful learning ability, it is in medical image segmentation (for example, thin under the cinereum matter segmentation of MR brain images, electron microscope The segmentation of after birth, the mitosis of mammary gland pathological image detection etc.) in achieved huge success.It is contemplated that by the framework For in the CNV segmentation task in OCT image.However, because OCT image feature is complex, directly use Convolutional neural networks carry out segmentation to CNV and two deficiencies be present：(1) traditional method is firstly the need of by image It is divided into some fritters (patch), is then based on patch training convolutional neural networks parted patterns.However, patch is with patch In the presence of some structural dependences (such as local similarity), traditional method in training convolutional neural networks parted pattern not In view of this effective structural information, so as to limit the segmentation performance of model.(2) traditional model is built upon single chi Spend on patch.The size of CNV is unfixed, so the patch of single yardstick is difficult to obtain effectively Contextual information, so as to have impact on segmentation precision.
The content of the invention
The technical problem to be solved by the invention is to provide a kind of OCT image median nexus film new vessels of can improving The OCT image median nexus film new vessels partitioning algorithm based on Multi-scale model priori convolutional neural networks of segmentation precision.
In order to solve the above technical problems, the technical solution adopted by the present invention is：
A kind of OCT image median nexus film new vessels partitioning algorithm comprises the following steps：
S01：A kind of structure priori learning method is designed to training image, builds structure prior matrix, the structure priori Matrix is used to distinguish CNV region and background area, comprises the following steps：
a：Super-pixel segmentation is carried out to training image and obtains some super-pixel；
b：Extract feature；
c：Each super-pixel is labeled as by 2 classes according to the correct mark of the training image, respectively choroid is new Angiogenic region and background area；
d：Use some super-pixel construction dictionaries marked；
e：All super-pixel are classified, obtain global structure priori figure；
f：Based on the global structure priori figure, local similar structure priori is calculated, tries to achieve the structure prior matrix；
S02：OCT original images are converted to by conspicuousness enhancing image based on the structure prior matrix, for strengthening train of thought The conspicuousness in film new vessels region；
S03：Multiscale analysis is used on conspicuousness enhancing image, conspicuousness enhancing image is divided into m Individual yardstick；
S04：Train to obtain the m convolutional neural networks models trained based on every kind of yardstick；
S05：Using step a, b, e, f in step S01 to test image carry out super-pixel segmentation, extraction feature, classification with And structure prior matrix is calculated, figure is carried out to test image based on the structure prior matrix obtained in S05 using step S02 As conversion, the test image after conversion is divided into m yardstick using step S03, utilizes the institute trained in step S04 State convolutional neural networks model to be tested, export m segmentation result, m segmentation is merged as final segmentation knot Fruit.
Super-pixel segmentation is carried out using SLIC algorithms.
The feature includes average gray value, the textural characteristics and part based on co-occurrence matrix of each super-pixel Gray feature.
Use K-means algorithm construction dictionaries, it is assumed that be concentrated with the data of N number of patient in training, each data include 2 points Class, K classes are polymerized to using K-means, then the data of N number of patient can be polymerized to 2KN classes altogether, obtain 2KN cluster centre, the cluster Center forms dictionary D, as shown in formula (1)：
D=[C1,1,C1,2…C1,K,B1,1,B1,2,..B1,K,…Cn,k..Bn,k…CN,1,..CN,K,BN,1,..BN,K] (1)
In formula, Cn,kRepresent k-th of cluster centre from CNV region of n-th of patient；Bn,kRepresent K-th of cluster centre from background area of n-th of patient, n=1,2 ... N, k=1,2 ... K.
Each super-pixel is classified using rarefaction representation, assorting process formalization is as shown in formula (2), formula In, x is desired sparse coefficient, and y is the super-pixel；
arg minx||x||1Subject to Dx=y (2)
Formula (2) is solved using SLEP tool boxes, obtains x solution；Point of the super-pixel is obtained using formula (3) Class result, x in formulaiRepresent the sparse coefficient of the i-th class, i=1,2 ... 2KN；
ri(y)=| | y-Dxi||2 (3)
2KN r is calculated according to formula (3)i(y), when r (y) value minimum, classification now is exactly the super picture The classification of element；All super-pixel classification to each image, you can obtain global space structure priori figure.
Based on the global space structure priori figure, local similar structure priori is calculated using Gaussian probability-density function, It is shown below：
M (a, b)=exp (- (cor (a, b)-c)2/u2) (4)
In formula, cor (a, b) is the coordinate each put in the global space structure priori figure, and c is the global space knot The center of structure priori figure median nexus film new vessels, radiuses of the u as the global space structure priori figure median nexus film, is root Tried to achieve according to center c and the average value of the distance of choroid boundary point, the matrix M tried to achieve is the structure prior matrix.
Image is changed using the structure prior matrix M, conversion formula such as formula (5)：
Is=MI0 (5)
In formula, I0It is original image, IsIt is conspicuousness enhancing image.
Fusion method described in step S05 is using maximum ballot criterion.
The beneficial effect that the present invention is reached：Structure prior matrix can obtain the structural dependence information between patch, The structure prior matrix in image is calculated using structure priori learning method, CNV area is come from available for enhancing Correlation between domain and the patch of background area, multiscale analysis can obtain effective contextual information of different scale, enter One step improves the segmentation precision of OCT image median nexus film new vessels.
Brief description of the drawings
Fig. 1 is the CNV segmentation flow chart based on Multi-scale model priori convolutional neural networks；
Fig. 2 is typical convolutional neural networks frame diagram；
Fig. 3 is CNV segmentation exemplary plot.
Embodiment
The invention will be further described below in conjunction with the accompanying drawings.Following examples are only used for clearly illustrating the present invention Technical scheme, and can not be limited the scope of the invention with this.
Embodiment 1
A kind of OCT image median nexus film new vessels partitioning algorithm, as shown in figure 1, this method mainly includes 2 stages： Training stage and test phase, are comprised the following steps that：
(1) training stage, mainly " study of structure priori " and " training of Multi-scale model priori convolutional neural networks " is included Two parts, comprise the following steps that：
S01：A kind of structure priori learning method is designed to training image, builds structure prior matrix, the structure priori Matrix is used to distinguish CNV region and background area；
The structure priori learning method comprises the following steps：
a：Super-pixel segmentation is carried out to training image and obtains some super-pixel regions, is split using SLIC algorithms；
b：Feature is extracted, the feature includes average gray value, the texture based on co-occurrence matrix of each super-pixel Feature and local gray level feature；
c：Each super-pixel is labeled as by 2 classes according to the correct mark of the training image, respectively choroid is new Angiogenic region and background area；
d：Using some super-pixel construction dictionaries marked, K-means algorithm construction dictionaries are used, it is assumed that instructing Practice the data for being concentrated with N number of patient, each data include 2 classification, are polymerized to K classes using K-means, then the data of N number of patient are total to 2KN classes can be polymerized to, obtain 2KN cluster centre, the cluster centre forms dictionary D, as shown in formula (1)：
D=[C1,1,C1,2…C1,K,B1,1,B1,2,..B1,K,…Cn,k..Bn,k…CN,1,..CN,K,BN,1,..BN,K] (1)
In formula, Cn,kRepresent k-th of cluster centre from CNV region of n-th of patient；Bn,kRepresent K-th of cluster centre from background area of n-th of patient；
e：All super-pixel are classified, obtain global structure priori, using rarefaction representation to each described super Pixel is classified, and assorting process formalization is as shown in formula (2), and in formula, x is desired sparse coefficient, and y is the super picture Element；
arg minx||x||1Subject to Dx=y (2)
Formula (2) is solved using SLEP tool boxes, obtains x solution；Point of the super-pixel is obtained using formula (3) Class result, x in formulaiRepresent the sparse coefficient of the i-th class, i=1,2 ... 2KN；
ri(y)=| | y-Dxi||2 (3)
2KN r is calculated according to formula (3)i(y), when r (y) value minimum, classification now is exactly the super picture The classification of element；All super-pixel classification to each image, you can obtain global space structure priori figure；
f：Based on the global structure priori, local similar structure priori is calculated using Gaussian probability-density function, is tried to achieve The structure prior matrix, local similar structure priori is calculated using local potential function, is shown below：
M (a, b)=exp (- (cor (a, b)-c)2/u2) (4)
In formula, cor (a, b) is the coordinate each put in the global space structure priori figure, and c is the global space knot The center of structure priori figure median nexus film new vessels, radiuses of the u as the global space structure priori figure median nexus film, is root Tried to achieve according to center c and the average value of the distance of choroid boundary point, the matrix M tried to achieve is structure prior matrix, and Fig. 1 gives Structure priori maps, it can be seen that the element in the regional area of the matrix has similar value from mapping, from the angle of the overall situation From the point of view of degree, the element in CNV and background area has dramatically different value, and therefore, the prior matrix should can For improving parted pattern for background area and the distinction of target area.
S02：OCT original images are converted to by conspicuousness enhancing image based on the structure prior matrix, for strengthening train of thought The conspicuousness in film new vessels region, image is changed using the structure prior matrix M, conversion formula such as formula (5)：
Is=MI0 (5)
In formula, I0It is original image, IsIt is conspicuousness enhancing image, as shown in figure 1, after image is changed, choroid is newborn The conspicuousness of blood vessel enhances really, and this is due to that conspicuousness enhancing image has incorporated structure prior information；
S03：Multiscale analysis is used on conspicuousness enhancing image, conspicuousness enhancing image is divided into m Individual yardstick；
S04：Train to obtain the m convolutional neural networks models trained based on every kind of yardstick, extracted for every kind of yardstick The patch of image, to the patch of every kind of yardstick, a structure priori convolutional neural networks model is trained, finally can then obtain m The individual structure priori convolutional neural networks model trained, each patch extracting method：Some pixel p using in image as Center, the square areas that a length of side is s is extracted, the class mark of pixel is the class mark of the patch, by using convolution god Patch is classified through network, so as to complete the classification to pixel, that is, completes segmentation task, the convolutional neural networks are adopted With the convolutional neural networks of classics, as shown in Fig. 2 in a convolutional neural networks, mainly by input layer, convolutional layer, Chi Hua Layer, full articulamentum and output layer composition.Convolutional layer typically has C convolution kernel, and convolution is done respectively to image, exports different reflect Penetrate.Convolutional layer can acquire the local features of different levels in image.In general, a pond layer can be added behind convolutional layer, The output of convolutional layer is the input of pond layer, and pond layer is typically down-sampled to input mapping progress using maximum pond method, that is, exists Point maximum in the neighborhood is selected to represent the neighborhood in one neighborhood.Pond layer can reduce the size of mapping, so as to reduce Computation complexity.After the circulation of convolutional layer several layers of below-pond layer, a full articulamentum is connected, the layer is by pond layer All output Mapping and Convertings be a column vector, an output layer is connected behind general one full articulamentum, output layer passes through One each input picture of softmax functions output belongs to the probability of each class, the conduct of the select probability maximum input picture Classification, the weight of convolutional neural networks is solved usually using stochastic gradient descent method；
(2) test phase
Super-pixel segmentation, extraction feature, classification and meter are carried out to test image using step a, b, e, f in step S01 Structure prior matrix is calculated, carrying out image to test image based on the structure prior matrix obtained in S05 using step S02 turns Change, the test image after conversion is divided into m yardstick using step S03, utilizes the volume trained in step S04 Product neural network model is tested, and is exported m segmentation result, m segmentation result is merged as final segmentation knot Fruit, the fusion method is using maximum ballot criterion.
(3) experimental result
The test of the inventive method has been carried out on 15 patient datas with CNV.Handed over using three foldings Verification method is pitched to examine the feasibility of this method and validity.
As shown in figure 3, giving the exemplary plot of 8 CNV segmentations, wherein red line represents that expert divides manually The borderline region of the CNV cut, white portion represents the region that the present invention is split automatically, by Fig. 3 segmentation result It is smaller with manual segmentation result error it can be seen that this method can effectively be partitioned into CNV region, using wearing this phase Like coefficient (DSC), the objective finger of True Positive Rate (TPVF), false positive rate (FPVF) and p value (p-values) as appraisal procedure Mark, the results are shown in Table 1 and table 2.
The Multi-scale model priori convolutional neural networks of table 1 and the segmentation performance based on sparse expression
The segmentation performance of the Multi-scale model priori convolutional neural networks of table 2 and convolutional neural networks
As a result show, be significantly better than using this method segmentation CNV based on rarefaction representation and convolutional Neural The segmentation effect of network.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art For member, without departing from the technical principles of the invention, some improvement and deformation can also be made, these are improved and deformation Also it should be regarded as protection scope of the present invention.

Claims (8)

1. a kind of OCT image median nexus film new vessels partitioning algorithm, it is characterized in that, comprise the following steps：
S01：A kind of structure priori learning method is designed to training image, builds structure prior matrix, the structure prior matrix For distinguishing CNV region and background area, comprise the following steps：
a：Super-pixel segmentation is carried out to training image and obtains some super-pixel；
b：Extract feature；
c：Each super-pixel is labeled as by 2 classes, the respectively new green blood of choroid according to the correct mark of the training image Area under control domain and background area；
d：Use some super-pixel construction dictionaries marked；
e：All super-pixel are classified, obtain global structure priori figure；
f：Based on the global structure priori figure, local similar structure priori is calculated, tries to achieve the structure prior matrix；
S02：OCT original images are converted to by conspicuousness enhancing image based on the structure prior matrix, it is new for strengthening choroid The conspicuousness in angiogenic region；
S03：Multiscale analysis is used on conspicuousness enhancing image, conspicuousness enhancing image is divided into m chi Degree；
S04：Train to obtain the m convolutional neural networks models trained based on every kind of yardstick；
S05：Super-pixel segmentation, extraction feature, classification are carried out to test image using step a, b, e, f in S01 and calculates knot Structure prior matrix, image conversion is carried out to test image based on the structure prior matrix obtained in S05 using S02, utilized The test image after conversion is divided into m yardstick by S03, utilizes the convolutional neural networks model trained in S04 Tested, export m segmentation result, m segmentation result is merged as final segmentation result.
2. a kind of OCT image median nexus film new vessels partitioning algorithm according to claim 1, it is characterized in that, use SLIC algorithms carry out super-pixel segmentation.
3. a kind of OCT image median nexus film new vessels partitioning algorithm according to claim 2, it is characterized in that, the spy Sign includes average gray value, the textural characteristics based on co-occurrence matrix and the local gray level feature of each super-pixel.
4. a kind of OCT image median nexus film new vessels partitioning algorithm according to claim 1, it is characterized in that, use K- Means algorithm construction dictionaries, it is assumed that be concentrated with the data of N number of patient in training, each data include 2 classification, use K-means Being polymerized to K classes, then the data of N number of patient can be polymerized to 2KN classes altogether, obtain 2KN cluster centre, and the cluster centre forms dictionary D, As shown in formula (1)：
D=[C1,1,C1,2…C1,K,B1,1,B1,2,..B1,K,…Cn,k..Bn,k…CN,1,..CN,K,BN,1,..BN,K] (1)
In formula, Cn,kRepresent k-th of cluster centre from CNV region of n-th of patient；Bn,kRepresent n-th K-th of cluster centre from background area of patient, n=1,2 ... N, k=1,2 ... K.
5. a kind of OCT image median nexus film new vessels partitioning algorithm according to claim 1, it is characterized in that, use is dilute Dredge and represent to classify to each super-pixel, assorting process formalization is as shown in formula (2), and in formula, x is to require sparse Coefficient, y are the super-pixel；
arg minx||x||1Subject to Dx=y (2)
Formula (2) is solved using SLEP tool boxes, obtains x solution；The classification knot of the super-pixel is obtained using formula (3) Fruit, x in formulaiRepresent the sparse coefficient of the i-th class, i=1,2 ... 2KN；
ri(y)=| | y-Dxi||2 (3)
2KN r is calculated according to formula (3)i(y), when r (y) value minimum, classification now is exactly the super-pixel Classification；All super-pixel classification to each image, you can obtain global space structure priori figure.
6. a kind of OCT image median nexus film new vessels partitioning algorithm according to claim 5, it is characterized in that, based on institute Global space structure priori figure is stated, local similar structure priori is calculated using Gaussian probability-density function, is shown below：
M (a, b)=exp (- (cor (a, b)-c)2/u2) (4)
In formula, cor (a, b) is the coordinate each put in the global space structure priori figure, and c is that the global space structure is first The center of figure median nexus film new vessels is tested, radiuses of the u as the global space structure priori figure median nexus film, is in Heart c and the average value of the distance of choroid boundary point are tried to achieve, and the matrix M tried to achieve is the structure prior matrix.
7. a kind of OCT image median nexus film new vessels partitioning algorithm according to claim 6, it is characterized in that, use institute State structure prior matrix M to change image, conversion formula such as formula (5)：
Is=MI0 (5)
In formula, I0It is original image, IsIt is conspicuousness enhancing image.
8. a kind of OCT image median nexus film new vessels partitioning algorithm according to claim 1, it is characterized in that, institute in S05 Fusion method is stated using maximum ballot criterion.
CN201710512064.3A 2017-06-28 2017-06-28 Choroid neogenesis blood vessel segmentation algorithm in OCT image Active CN107369160B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201710512064.3A CN107369160B (en) 2017-06-28 2017-06-28 Choroid neogenesis blood vessel segmentation algorithm in OCT image
PCT/CN2018/089056 WO2019001208A1 (en) 2017-06-28 2018-05-30 Segmentation algorithm for choroidal neovascularization in oct image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710512064.3A CN107369160B (en) 2017-06-28 2017-06-28 Choroid neogenesis blood vessel segmentation algorithm in OCT image

Publications (2)

Publication Number Publication Date
CN107369160A true CN107369160A (en) 2017-11-21
CN107369160B CN107369160B (en) 2020-04-03

Family

ID=60305691

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710512064.3A Active CN107369160B (en) 2017-06-28 2017-06-28 Choroid neogenesis blood vessel segmentation algorithm in OCT image

Country Status (2)

Country Link
CN (1) CN107369160B (en)
WO (1) WO2019001208A1 (en)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107945176A (en) * 2017-12-15 2018-04-20 西安中科微光影像技术有限公司 A kind of colour IVOCT imaging methods
CN108198184A (en) * 2018-01-09 2018-06-22 北京理工大学 The method and system of contrastographic picture medium vessels segmentation
CN108550136A (en) * 2018-03-14 2018-09-18 宁波大学 A kind of eye fundus image blood vessel segmentation method
CN108875555A (en) * 2018-04-25 2018-11-23 中国人民解放军军事科学院军事医学研究院 Video interest neural network based region and well-marked target extraction and positioning system
CN108961280A (en) * 2018-06-29 2018-12-07 电子科技大学 A kind of eyeground optic disk fine segmentation method based on SLIC super-pixel segmentation
CN109035252A (en) * 2018-06-29 2018-12-18 山东财经大学 A kind of super-pixel method towards medical image segmentation
WO2019001208A1 (en) * 2017-06-28 2019-01-03 苏州比格威医疗科技有限公司 Segmentation algorithm for choroidal neovascularization in oct image
CN109741335A (en) * 2018-11-28 2019-05-10 北京理工大学 Blood vessel OCT image medium vessels wall and the dividing method and device of blood flow area
WO2019128971A1 (en) * 2017-12-26 2019-07-04 云象科技股份有限公司 Control method for automated microscope system, microscope system and computer-readable storage medium
CN110415245A (en) * 2019-08-05 2019-11-05 上海鹰瞳医疗科技有限公司 Optical data determines method, model training method and equipment
CN110415231A (en) * 2019-07-25 2019-11-05 山东浪潮人工智能研究院有限公司 A kind of CNV dividing method based on attention pro-active network
CN110555853A (en) * 2019-08-07 2019-12-10 杭州深睿博联科技有限公司 Method and device for segmentation algorithm evaluation based on anatomical priors
CN111767929A (en) * 2019-03-14 2020-10-13 上海市第一人民医院 Method and system for constructing sub-macular neovascularization model
CN112365452A (en) * 2020-10-26 2021-02-12 江苏恒力化纤股份有限公司 Network wire network point detection method based on bilateral images
WO2022127043A1 (en) * 2020-12-15 2022-06-23 平安科技（深圳）有限公司 Detection method and apparatus based on convolutional neural network, and computer device and storage medium

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109829507B (en) * 2019-02-21 2023-09-19 国网上海市电力公司 Aerial high-voltage transmission line environment detection method
CN110309827B (en) * 2019-05-06 2022-10-14 上海海洋大学 Edema region segmentation model based on OCT image
CN110163213B (en) * 2019-05-16 2022-12-06 西安电子科技大学 Remote sensing image segmentation method based on disparity map and multi-scale depth network model
CN110472639B (en) * 2019-08-05 2023-04-18 山东工商学院 Target extraction method based on significance prior information
CN112686902B (en) * 2019-10-17 2023-02-03 西安邮电大学 Two-stage calculation method for brain glioma identification and segmentation in nuclear magnetic resonance image
CN110991463B (en) * 2019-11-04 2023-05-02 同济大学 Multi-scale guide filtering feature extraction method under guidance of super-pixel map
CN111028210B (en) * 2019-11-25 2023-07-18 北京航天控制仪器研究所 Glass tube end face defect detection method based on deep neural network
CN111325729A (en) * 2020-02-19 2020-06-23 青岛海信医疗设备股份有限公司 Biological tissue segmentation method based on biomedical images and communication terminal
CN111709317B (en) * 2020-05-28 2023-08-04 深圳市旭景数字技术有限公司 Pedestrian re-identification method based on multi-scale features under saliency model
CN111523521B (en) * 2020-06-18 2023-04-07 西安电子科技大学 Remote sensing image classification method for double-branch fusion multi-scale attention neural network
CN111860465A (en) * 2020-08-10 2020-10-30 华侨大学 Remote sensing image extraction method, device, equipment and storage medium based on super pixels
CN113313164A (en) * 2021-05-27 2021-08-27 复旦大学附属肿瘤医院 Digital pathological image classification method and system based on superpixel segmentation and image convolution
CN116503426B (en) * 2023-06-26 2023-08-25 天津大学 Ultrasonic image segmentation method based on image processing

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150186771A1 (en) * 2012-09-10 2015-07-02 FEM, Inc. Prospective media content generation using neural network modeling
WO2015105239A1 (en) * 2014-01-13 2015-07-16 삼성테크윈 주식회사 Vehicle and lane position detection system and method
CN105825502A (en) * 2016-03-12 2016-08-03 浙江大学 Saliency-guidance-based weak supervision image analysis method of dictionary learning
CN105931255A (en) * 2016-05-18 2016-09-07 天津工业大学 Method for locating target in image based on obviousness and deep convolutional neural network
CN106127164A (en) * 2016-06-29 2016-11-16 北京智芯原动科技有限公司 The pedestrian detection method with convolutional neural networks and device is detected based on significance
CN106157319A (en) * 2016-07-28 2016-11-23 哈尔滨工业大学 The significance detection method that region based on convolutional neural networks and Pixel-level merge
CN106446914A (en) * 2016-09-28 2017-02-22 天津工业大学 Road detection based on superpixels and convolution neural network

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106599804B (en) * 2016-11-30 2019-07-05 哈尔滨工业大学 Fovea centralis detection method based on multiple features model
CN107369160B (en) * 2017-06-28 2020-04-03 苏州比格威医疗科技有限公司 Choroid neogenesis blood vessel segmentation algorithm in OCT image

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150186771A1 (en) * 2012-09-10 2015-07-02 FEM, Inc. Prospective media content generation using neural network modeling
WO2015105239A1 (en) * 2014-01-13 2015-07-16 삼성테크윈 주식회사 Vehicle and lane position detection system and method
CN105825502A (en) * 2016-03-12 2016-08-03 浙江大学 Saliency-guidance-based weak supervision image analysis method of dictionary learning
CN105931255A (en) * 2016-05-18 2016-09-07 天津工业大学 Method for locating target in image based on obviousness and deep convolutional neural network
CN106127164A (en) * 2016-06-29 2016-11-16 北京智芯原动科技有限公司 The pedestrian detection method with convolutional neural networks and device is detected based on significance
CN106157319A (en) * 2016-07-28 2016-11-23 哈尔滨工业大学 The significance detection method that region based on convolutional neural networks and Pixel-level merge
CN106446914A (en) * 2016-09-28 2017-02-22 天津工业大学 Road detection based on superpixels and convolution neural network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
GAYOUNG LEE 等: "deep saliency with Encoded Low Level Distance Map and High Level Features", 《THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 *
李岳云 等: "深度卷积神经网络的显著性检测", 《中国图象图形学报》 *
饶倩 等: "基于超像素的人工神经网络图像分类", 《计算机与现代化》 *

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019001208A1 (en) * 2017-06-28 2019-01-03 苏州比格威医疗科技有限公司 Segmentation algorithm for choroidal neovascularization in oct image
CN107945176A (en) * 2017-12-15 2018-04-20 西安中科微光影像技术有限公司 A kind of colour IVOCT imaging methods
CN107945176B (en) * 2017-12-15 2021-05-11 西安中科微光影像技术有限公司 Color IVOCT imaging method
WO2019128971A1 (en) * 2017-12-26 2019-07-04 云象科技股份有限公司 Control method for automated microscope system, microscope system and computer-readable storage medium
CN108198184B (en) * 2018-01-09 2020-05-05 北京理工大学 Method and system for vessel segmentation in contrast images
CN108198184A (en) * 2018-01-09 2018-06-22 北京理工大学 The method and system of contrastographic picture medium vessels segmentation
CN108550136A (en) * 2018-03-14 2018-09-18 宁波大学 A kind of eye fundus image blood vessel segmentation method
CN108875555A (en) * 2018-04-25 2018-11-23 中国人民解放军军事科学院军事医学研究院 Video interest neural network based region and well-marked target extraction and positioning system
CN108875555B (en) * 2018-04-25 2022-02-25 中国人民解放军军事科学院军事医学研究院 Video interest area and salient object extracting and positioning system based on neural network
CN109035252A (en) * 2018-06-29 2018-12-18 山东财经大学 A kind of super-pixel method towards medical image segmentation
CN108961280B (en) * 2018-06-29 2021-11-02 电子科技大学 Fundus optic disc fine segmentation method based on SLIC super-pixel segmentation
CN108961280A (en) * 2018-06-29 2018-12-07 电子科技大学 A kind of eyeground optic disk fine segmentation method based on SLIC super-pixel segmentation
CN109741335A (en) * 2018-11-28 2019-05-10 北京理工大学 Blood vessel OCT image medium vessels wall and the dividing method and device of blood flow area
CN111767929A (en) * 2019-03-14 2020-10-13 上海市第一人民医院 Method and system for constructing sub-macular neovascularization model
CN110415231A (en) * 2019-07-25 2019-11-05 山东浪潮人工智能研究院有限公司 A kind of CNV dividing method based on attention pro-active network
CN110415245A (en) * 2019-08-05 2019-11-05 上海鹰瞳医疗科技有限公司 Optical data determines method, model training method and equipment
CN110555853A (en) * 2019-08-07 2019-12-10 杭州深睿博联科技有限公司 Method and device for segmentation algorithm evaluation based on anatomical priors
CN110555853B (en) * 2019-08-07 2022-07-19 杭州深睿博联科技有限公司 Method and device for segmentation algorithm evaluation based on anatomical priors
CN112365452A (en) * 2020-10-26 2021-02-12 江苏恒力化纤股份有限公司 Network wire network point detection method based on bilateral images
CN112365452B (en) * 2020-10-26 2023-06-02 江苏恒力化纤股份有限公司 Network wire network point detection method based on double-side image
WO2022127043A1 (en) * 2020-12-15 2022-06-23 平安科技（深圳）有限公司 Detection method and apparatus based on convolutional neural network, and computer device and storage medium

Also Published As

Publication number Publication date
WO2019001208A1 (en) 2019-01-03
CN107369160B (en) 2020-04-03

Similar Documents

Publication Publication Date Title
CN107369160A (en) A kind of OCT image median nexus film new vessels partitioning algorithm
CN106056595B (en) Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules
CN108257135A (en) The assistant diagnosis system of medical image features is understood based on deep learning method
WO2021088747A1 (en) Deep-learning-based method for predicting morphological change of liver tumor after ablation
US20210035689A1 (en) Modeling method and apparatus for diagnosing ophthalmic disease based on artificial intelligence, and storage medium
CN106530283A (en) SVM (support vector machine)-based medical image blood vessel recognition method
CN107506770A (en) Diabetic retinopathy eye-ground photography standard picture generation method
CN109858540B (en) Medical image recognition system and method based on multi-mode fusion
CN109493954A (en) A kind of SD-OCT image retinopathy detection system differentiating positioning based on classification
CN110310281A (en) Lung neoplasm detection and dividing method in a kind of Virtual Medical based on Mask-RCNN deep learning
WO2022001571A1 (en) Computing method based on super-pixel image similarity
Panda et al. New binary Hausdorff symmetry measure based seeded region growing for retinal vessel segmentation
CN109118495A (en) A kind of Segmentation Method of Retinal Blood Vessels and device
CN109447998A (en) Based on the automatic division method under PCANet deep learning model
CN109214397A (en) The dividing method of Lung neoplasm in a kind of lung CT image
CN109636805A (en) A kind of uterine neck image lesion region segmenting device and method based on classification priori
CN106651874B (en) Space domain splitting method after brain tumor surgery based on multi-modal MRI data
CN109300113A (en) A kind of Lung neoplasm assisted detection system and method based on improvement Convex Hull Method
CN106529188A (en) Image processing method applied to surgical navigation
Fan et al. Lung nodule detection based on 3D convolutional neural networks
CN109035227A (en) The system that lung tumors detection and diagnosis is carried out to CT image
CN107229937A (en) A kind of retinal vessel sorting technique and device
CN111767952A (en) Interpretable classification method for benign and malignant pulmonary nodules
Sreelakshmy et al. An automated deep learning model for the cerebellum segmentation from fetal brain images
David et al. Retinal blood vessels and optic disc segmentation using U-net

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
PE01 Entry into force of the registration of the contract for pledge of patent right
PE01 Entry into force of the registration of the contract for pledge of patent right

Denomination of invention: A segmentation algorithm of choroidal neovascularization in OCT images

Effective date of registration: 20220520

Granted publication date: 20200403

Pledgee: Suzhou high tech Industrial Development Zone sub branch of Bank of Communications Co.,Ltd.

Pledgor: SUZHOU BIGVISION MEDICAL TECHNOLOGY Co.,Ltd.

Registration number: Y2022320010152