CN104573712B - Arteriovenous retinal vessel sorting technique based on eye fundus image - Google Patents
Arteriovenous retinal vessel sorting technique based on eye fundus image Download PDFInfo
- Publication number
- CN104573712B CN104573712B CN201410850207.8A CN201410850207A CN104573712B CN 104573712 B CN104573712 B CN 104573712B CN 201410850207 A CN201410850207 A CN 201410850207A CN 104573712 B CN104573712 B CN 104573712B
- Authority
- CN
- China
- Prior art keywords
- blood vessel
- main
- fundus image
- eye fundus
- vessel
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 58
- 210000001210 retinal vessel Anatomy 0.000 title claims abstract description 17
- 210000004204 blood vessel Anatomy 0.000 claims abstract description 241
- 210000003733 optic disk Anatomy 0.000 claims abstract description 40
- 230000002792 vascular Effects 0.000 claims description 43
- 230000011218 segmentation Effects 0.000 claims description 41
- 239000013598 vector Substances 0.000 claims description 27
- 210000003462 vein Anatomy 0.000 claims description 10
- 238000006243 chemical reaction Methods 0.000 claims description 8
- 238000004040 coloring Methods 0.000 claims description 7
- 239000012141 concentrate Substances 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 5
- 210000001367 artery Anatomy 0.000 claims description 4
- 239000012634 fragment Substances 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000009792 diffusion process Methods 0.000 abstract description 2
- 239000008280 blood Substances 0.000 description 11
- 210000004369 blood Anatomy 0.000 description 11
- 125000004122 cyclic group Chemical group 0.000 description 11
- 230000003902 lesion Effects 0.000 description 9
- 208000032843 Hemorrhage Diseases 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000000740 bleeding effect Effects 0.000 description 4
- 238000004195 computer-aided diagnosis Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000003745 diagnosis Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 230000002207 retinal effect Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000002577 ophthalmoscopy Methods 0.000 description 2
- 206010000234 Abortion spontaneous Diseases 0.000 description 1
- 206010008190 Cerebrovascular accident Diseases 0.000 description 1
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 229920000742 Cotton Polymers 0.000 description 1
- 208000006011 Stroke Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000002490 cerebral effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 210000000416 exudates and transudate Anatomy 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010348 incorporation Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 208000015994 miscarriage Diseases 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000009659 non-destructive testing Methods 0.000 description 1
- 210000001328 optic nerve Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 208000000995 spontaneous abortion Diseases 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000002604 ultrasonography 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
- G06F18/23—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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/457—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by analysing connectivity, e.g. edge linking, connected component analysis or slices
Landscapes
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Heart & Thoracic Surgery (AREA)
- Probability & Statistics with Applications (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- General Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Ophthalmology & Optometry (AREA)
- Biophysics (AREA)
- Multimedia (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The invention discloses a kind of arteriovenous retinal vessel sorting technique based on eye fundus image, including：(1) the global blood vessel collection and optic disk location information of eye fundus image are obtained, described global blood vessel collection is the set of all blood vessels in the eye fundus image, and described optic disk location information includes the optic disk center of the eye fundus image；(2) main blood vessel is determined according to described global blood vessel collection and optic disk location information, and main blood vessel is classified to obtain main blood vessel classification information；(3) blood vessel concentrated based on SAT breadth first search method to described global blood vessel is used to be classified to obtain global classification information using described main blood vessel classification information.The present invention obtains the classification information of the main blood vessel around optic disk first, and the breadth first search method for starting based on from main blood vessel SAT extends out diffusion and obtains all blood vessels, realizes a complete automatic blood vessel sorting technique, without manual intervention, and nicety of grading is high.
Description
Technical field
The present invention relates to computer-aided diagnosis technical field, and in particular to a kind of arteriovenous view based on eye fundus image
Film blood vessel sorting technique.
Background technology
With the fast development of the artificial intelligence field in computer technology, computer-aided diagnosis technology is also gradually sent out
Exhibition.Computer-aided diagnosis technology refers to by iconography, Medical Image Processing and other possible physiology, biochemical hand
Section, calculated with reference to the analysis of computer, assisted image section doctor has found focus, improves the accuracy rate of diagnosis.
Usual Medical Imaging Computer auxiliary diagnosis is divided into three steps, specific as follows：The first step is from normal lesion
Extracted in structure；Second step is the quantization of characteristics of image；3rd step is that data are handled and drawn a conclusion.
Because computer can carry out accurate quantitative calculating using image information comprehensively, the subjectivity of people is removed, is avoided
The diagnostic result " to be varied " because of caused by the difference of personal knowledge and experience；So its result is free from paste, it is
Determine, it makes diagnosis become more accurate, more science.
With the development of modern high technology, computer-aided diagnosis will with the technological incorporation such as image procossing and PACS system,
Become easier to operation, also more they tend to accurately, its clinical application range will further expand.
In medical science detection, eyes are uniquely can Non-Destructive Testing while informative organ.Retinal blood is pointed out in research
Blood vessel in pipe lesion limits to constriction, diffuses constriction, arteriovenous crossing compression, blood vessel walking change, copper wire artery, bleeding, cotton
Wadding spot, hard exudate and retinal nerve fibre layer defect have significant correlation with brain soldier.And for the prediction of cerebral apoplexy,
Funduscopy only needs 40 yuan, and MRI inspections then need thousands of members, and carotid ultrasound is also required to 140 yuan.Funduscopy by contrast
Cost performance highest.The full-automatic method of eye fundus image computer analysis, including instant PVR can be provided
Classification, without expert opinion, establish and predict that the system of three high concurrent diseases has its certain warp with optical fundus blood vessel optic nerve
Ji meaning.Therefore, the lesion detection of retinal vessel has outstanding role in the auxiliary detection to brain soldier.One is wherein built to move
The automatic checkout system of vein crossings compressing retinal vasculopathy is even more key component therein.
The disease that blood vessel segmentation, optic disk positioning and blood vessel classification (arteriovenous division) are retinal vessels is carried out to eye fundus image
Become the basis of detection, existing blood vessel segmentation method needs manually to add markup information, and automaticity is not high.
The content of the invention
In view of the shortcomings of the prior art, the invention provides a kind of arteriovenous retinal vessel classification based on eye fundus image
Method.
A kind of arteriovenous retinal vessel sorting technique based on eye fundus image, comprises the following steps：
(1) eye fundus image is directed to, several different global blood vessel collection are obtained based on blood vessel segmentation method；
(2) for each global blood vessel collection, based on final convergence region corresponding to the acquisition of fuzzy convergence algorithm, and according to each
The overlapping cases of individual final convergence region determine the optic disk center of the eye fundus image；
(3) a global blood vessel collection is selected, main blood vessel is determined according to the global blood vessel collection and optic disk location information, and to master
Blood vessel is classified to obtain main blood vessel classification information；
(4) described main blood vessel classification information is utilized to use based on SAT breadth first search method to described global blood vessel
The blood vessel of concentration is classified to obtain global classification information.
Selection corresponds to the global blood vessel collection of the pixel quantity of blood vessel at most, corresponding step in step (3) in the present invention
Suddenly the global blood vessel collection of the selection is classified in (4).
The purpose that the present invention obtains several different global blood vessel collection is to determine convergence by multiple global blood vessel collection
Region, optic disk center is obtained according to the overlapping cases between convergence region corresponding to each global blood vessel collection, is advantageously ensured that
The accuracy at the optic disk center arrived.
It is first when the step (1) obtains several different global blood vessel collection as follows based on blood vessel segmentation method
Several different binary-state thresholds are first set, are then proceeded as follows for each binary-state threshold：
(1-1) carries out binary conversion treatment according to current binary-state threshold to described eye fundus image, and extracts at binaryzation
The center line and edge in eye fundus image after reason, obtains vascular tree；
(1-2) does disconnection process to described vascular tree crotch and obtains vessel segment, and enters line point to each vessel segment
Cut to obtain blood vessel, obtain primitive vessel collection；
(1-3) determines segmentation blood vessel by mistake, and concentrates to remove from primitive vessel and obtain global blood vessel collection.
It after binary conversion treatment is that the pixel number of blood vessel accounts for the picture of whole eye fundus image that binary-state threshold, which is, in the present invention
Vegetarian refreshments ratio, usual value are 4~20%.Binary-state threshold is bigger, then looser.Preferably, described binary-state threshold
For 14%.
Mistake segmentation blood vessel in the step (1-3) includes being divided into two classes, and segmentation blood vessel is based on blood vessel both sides to the first kind by mistake
Background difference determine, the second class be based on shape of blood vessel determination.
The difference of heretofore described background difference value background color, is determined as follows in step (1-3)
One kind splits blood vessel by mistake：
(a1) each blood vessel is directed to, extracts the characteristic vector of the blood vessel both sides background；
The characteristic vector of any one side back scape is according to 5~8 pixels of the lateral extent center line with all in inner region
Color on tri- passages of RGB of pixel (being less than 5~10 pixels with the distance of center line) is worth to.
Be a three-dimensional vector per the characteristic vector of side, respectively expression blood vessel both sides background on tri- passages of RGB
Color value information.The lateral extent center line is obtained during specific implementation less than all pixels point in the region of 5~10 pixels
Color value on tri- passages of R, G, B is simultaneously averaging on each passage respectively, and then obtains the characteristic vector of the side.
(a2) clustering procedure is used to gather all blood vessels for two classes according to the characteristic vector, obtained group is optic disk week
Enclose segmentation blood vessel by mistake.
For each blood vessel, the Euclidean distance of calculating both sides characteristic vector, then to Euclidean distance corresponding to all blood vessels
Clustered, that is, complete the cluster to blood vessel.
Because K mean cluster (i.e. K-means algorithms) does not need adjusting parameter, and the speed of service is very fast.Preferably, institute
Stating in step (a2) uses K mean cluster method to gather all blood vessels for two classes.
Heretofore described shape of blood vessel actually refers to the annexation of each blood vessel, by as follows in step (1-3)
Step determines the second class segmentation blood vessel by mistake：
It is determined that the cyclic structure in marking off the eye fundus image of primitive vessel collection, for each cyclic structure, if the ring-type
The length of the maximum blood vessel of length is less than default segmentation length threshold in structure, then it is assumed that all blood vessels are equal in the cyclic structure
Split blood vessel by mistake for the second class, further proceed as follows：
The center of the cyclic structure is determined, and calculates the center to length more than or equal to the blood vessel of segmentation length threshold
Beeline (i.e. distance of the center to the length away from its nearest neighbours more than or equal to the blood vessel of segmentation length threshold), it is believed that with
The center is the center of circle, beeline is that all blood vessels are the second class segmentation blood vessel by mistake in the border circular areas of radius.
The setting for splitting length threshold is set according to practical experience value, ring-type segmentation length threshold α=x/ in the present invention
60~x/45.
The blood vessel segmentation method of the present invention, further determine to divide by mistake using the background and shape of blood vessel after obtaining primitive vessel
The blood vessel cut, can effectively it remove because ring-type is reflective caused by taking pictures, the jump rank edge of non-vascular around optic disk, plaque-like disease
Split blood vessel by mistake caused by the reason such as change and bleeding lesion, substantially increase vessel segmentation (the global blood vessel obtained
Collection) accuracy.
Proceeded as follows when convergence region corresponding to current global blood vessel collection is obtained in the step (2)：
(2-1) concentrates each blood vessel to current global blood vessel, and the region of convergence of the blood vessel is obtained using fuzzy convergence algorithm
Domain；
Ballot of the number of convergence region corresponding to (2-2) statistics each pixel of eye fundus image as the pixel
Value；
(2-3) chooses the big preceding n pixel of ballot value, and the region based on eight connections is used to n pixel of selection
Connection algorithm obtains several connected regions, and is used as final convergence region using the maximum connected region of area.
N size is set according to the size of eye fundus image, preferably, the value of the n is 1000~3000.
The step (2) determines the eye fundus image according to the overlapping cases of each final convergence region by the following method
Optic disk center：
The overlapping region of at least l final convergence regions is judged whether, wherein l=k/3~k/2, K are binaryzation threshold
The number of value：
If in the presence of the centre coordinate of the overlapping region maximum using area is used as optic disk location information；
Otherwise, optic disk location information is obtained with use specific template matching method.
The step (3) determines main blood vessel by the following method：
Using apart from optic disk center be the region within several pixels as optic disk adjacent domain, it is adjacent with described optic disk
Length is more than the blood vessel of default classification length threshold as main blood vessel near field.
In the present invention using apart from optic disk center as the region within R pixel, i.e., using optic disk center as the center of circle, using R as
The region of radius is more than default classification length threshold as optic disk adjacent domain with length in the optic disk adjacent domain of determination
Blood vessel is as main blood vessel.
Wherein, the size of radius R and classification length threshold determines according to the size and actual conditions of eyeground picture.As excellent
Choosing, the value of the R is 100~150, and described classification length threshold is 50~65.
The step (3) is classified to obtain main blood vessel classification information as follows to main blood vessel：
(3-1) obtains the average caliber of each main blood vessel, and it is vein blood vessel to specify the maximum main blood vessel of average caliber；
Each main blood vessel is cut into some fragments by (3-2), obtains corresponding main Vascular Slice；
(3-3) extracts the characteristic vector of each main Vascular Slice, and uses clustering procedure by institute based on described characteristic vector
The main Vascular Slice stated is gathered for two classes, and using the class where by main Vascular Slice corresponding to vein blood vessel as vein blood vessel, separately
One kind is used as arteries；
(3-4) is directed to each main blood vessel, and the classification results of the main blood vessel are used as using the class where more main Vascular Slice.
The characteristic vector of each main Vascular Slice is extracted in the step (3-3) by the following method：
Obtain the color of all pixels point in the region within the blood vessel center of main Vascular Slice several pixels
Information, and the characteristic vector of the main Vascular Slice is used as using the average of the colouring information of all pixels point in the region.
Preferably, the blood vessel center 5 apart from main Vascular Slice is obtained during the characteristic vector of each main Vascular Slice of extraction
The colouring information of all pixels point in region within~8 pixels.
Breadth first search is carried out to global blood vessel collection, three treaties are used in search procedure based on SAT breadth first search method
Beam condition carries out the transmission of blood vessel classification：Constraints 1：Two blood vessels of right-angled intersection are respectively labeled as two class blood vessels；Constraint
Condition 2：Three trouble structures in three vascular markers be；Constraints 3：Three blood vessels wherein blood vessel in three trouble structures is such as
When fruit and remaining two blood vessel angles sum are less than or equal to 270 degree, be determined as three trouble structures of constraints 2, i.e., three
Blood vessel is same class blood vessel, does not otherwise do and judges.
When can preferably be distinguished due to blood vessel segmentation using above three constraints it is existing leakage segmentation and caused by
Intersect the situation that erroneous judgement is broken into trident, improve nicety of grading.On the other hand, first obtained based on constraints blood vessel with a high credibility
To classification results, the mode to make corrections is done in the region that blood vessel with a low credibility is not delivered to for blood vessel with a high credibility to be caused
The confidence level of global blood vessel mark improves, so as to improve classifying quality.
The present invention is existing, and to specify thicker main blood vessel be vein blood vessel, therefore the blood in whole SAT breadth first search method
Pipe is thicker, and the confidence level of the classification results of the obtained blood vessel is higher.
Compared with prior art, the present invention obtains the classification information of the main blood vessel around optic disk first, and from main blood vessels open
Primordium extends out diffusion in SAT breadth first search method and obtains all blood vessels, realizes a complete automatic blood vessel sorting technique,
Without manual intervention, and nicety of grading is high.
Brief description of the drawings
Fig. 1 is the eye fundus image of the present embodiment；
Fig. 2 is the flow chart that the arteriovenous retinal vessel of the eye fundus image based on breadth first search method is classified；
Fig. 3 is the flow chart for carrying out blood vessel segmentation in the present embodiment to eye fundus image；
Fig. 4 is the schematic diagram for the primitive vessel collection that blood vessel segmentation obtains；
Fig. 5 is the schematic diagram for the global blood vessel collection that blood vessel segmentation obtains；
The flow chart of optic disk positioning is carried out in Fig. 6 the present embodiment to eye fundus image；
Fig. 7 is the schematic diagram for the global classification information that the arteriovenous retinal vessel of the present embodiment eye fundus image is classified.
Embodiment
Below in conjunction with the drawings and specific embodiments, the present invention will be described in detail.
The present embodiment illustrates the arteriovenous retinal vessel point based on eye fundus image by taking the eye fundus image shown in Fig. 1 as an example
Class method, the size of the eye fundus image is 3000 × 3000.The ring-type as caused by taking pictures is reflective, the jump of non-vascular around optic disk
The reasons such as rank edge, plaque-like lesion and bleeding lesion, bright ring be present in the eye fundus image.
Arteriovenous retinal vessel classification is carried out to the eye fundus image, classification process is as shown in Fig. 2 comprise the following steps：
(1) obtain the global blood vessel collection (i.e. final blood vessel collection) of eye fundus image and optic disk location information, global blood vessel collection are
The set of all blood vessels in eye fundus image, optic disk location information include the optic disk center of eye fundus image；
The global blood vessel collection for obtaining eye fundus image by carrying out blood vessel segmentation to eye fundus image in the present embodiment, idiographic flow
As shown in figure 3, comprise the following steps：
(1-1) carries out wavelet transformation (IUWT small echos) to eye fundus image, according to default binary-state threshold to by small echo
The eye fundus image of conversion carries out binary conversion treatment, and extracts center line and edge in the eye fundus image after binary conversion treatment, obtains
To vascular tree；
(1-2) does disconnection process to vascular tree crotch and obtains vessel segment, and enters line to each vessel segment and split to obtain
Blood vessel, obtain primitive vessel collection.
When doing disconnection process to vascular tree crotch：When more center lines are pooled to one in the vessel centerline in vascular tree
During point, central point (crosspoint collected) is removed, obtains single more vessel centerlines.
When entering line segmentation to each vessel segment：One vessel segment is used as using each center line.Vessel segment is a song
Line, the conventional method split with the line of image procossing, by curve with more beeline approachings.More obtained straight lines, every straight
Line represents a blood vessel, and the set of all straight lines is primitive vessel collection.
(1-3) determines segmentation blood vessel by mistake, splits blood vessel in the present embodiment by mistake and obtains the first kind segmentation blood vessel and the second class by mistake
Segmentation blood vessel by mistake, the first kind is deleted from primitive vessel set and splits blood vessel and the second class segmentation blood vessel by mistake by mistake, then obtains the overall situation
Blood vessel collection (i.e. final blood vessel collection).
For ring-type it is reflective caused by split by mistake, it is blood by segment that its blood vessel being partitioned into is relative to be had with normal blood vessels
The design feature of the ring of pipe composition.
Split by mistake for caused by the jump rank edge around optic disk, the blood vessel that it is partitioned into is in rgb color space and structure
The characteristics of upper not special.Segmentation blood vessel is that the background around optic disk forms by mistake for it, because it is close to optic disk, and around optic disk
Background color be relatively distant from for the background around optic disk and conventional vascular color have acquaintance property；From a structural point due to
It is isolated presence, and optic disk peripheral vessels is mixed in together also is difficult to distinguish from structure, if done by force from structure
Judgement easily causes substantial amounts of erroneous judgement.But the background of its vessel segment both sides on rgb color space for there is larger color
Difference, because its both sides background generic background while by optic disk and in addition while be made up of.Compared in general blood vessel, its both sides
Background is all made up of by generic background or all optic disk.
Split by mistake caused by becoming for plaque-like lesion and hemorrhage, its blood vessel being partitioned into is by commonly carrying on the back in color
Scape forms, without special characteristic.But its structure is with respect to seeming especially mixed and disorderly for normal blood vessels, without longer blood vessel
The tree of formation, mostly multiple small cyclic structures and some thin vessels in small, broken bits combine.
Analyzed based on more than, the background difference based on blood vessel both sides determines first kind segmentation blood vessel by mistake in the present embodiment：
(a1) each blood vessel is directed to, extracts the characteristic vector of the blood vessel both sides background；
The all pixels point in the region of 10 pixels of the lateral extent center line etc is obtained on tri- passages of R, G, B
Color value and be averaging respectively on each passage, and then obtain the characteristic vector of the side.
Characteristic vector per side is actually a three-dimensional vector, represents the logical in RGB tri- of blood vessel both sides background respectively
Color value information on road.
(a2) K mean cluster method is used to gather characteristic vector for two classes, will according to the corresponding relation of characteristic vector and blood vessel
All blood vessels are divided into two classes, because probability of miscarriage of justice generally will not be too high, therefore obtained group (the i.e. less blood of blood vessel content
Pipe) it is first kind segmentation blood vessel by mistake.
By determining the second class segmentation blood vessel by mistake based on shape of blood vessel in the present embodiment：
(b1) determine to mark off the cyclic structure in the eye fundus image of primitive vessel collection.
Non-directed graph G=(V, E) can be built during specific implementation, V is the set of two end points of all vessel centerlines, E
For the set of the center line of all blood vessels, cyclic structure is determined using non-directed graph G=(V, E).
(b2) each cyclic structure is directed to, if the length of the maximum blood vessel of length is less than default segmentation in the cyclic structure
Length threshold α, wherein α=x/60~x/45, (it is the transverse direction of eye fundus image to split length threshold α=x/50, x in the present embodiment
Size, i.e. x=3000), then blood vessel all in the cyclic structure is the second class segmentation blood vessel by mistake, is further grasped as follows
Make：
The center of the cyclic structure is determined, and calculates the beeline of blood vessel of the center to length more than or equal to α (i.e.
The distance of blood vessel of the center to the length away from its nearest neighbours more than or equal to α), it is half by the center of circle, beeline of the center
All blood vessels are the second class segmentation blood vessel by mistake in the border circular areas in footpath.
It is that the pixel number of blood vessel accounts for whole eye fundus image after binary conversion treatment that binary-state threshold, which is, in the present embodiment
Pixel ratio, usual value are 4~20%.Binary-state threshold is bigger, then looser.
Six different binary-state thresholds, respectively 4%, 6%, 8%, 10%, 12% and 14% are used in the present embodiment.
Step (1-1)~(1-3) is carried out for each binary-state threshold, corresponds to 6 global blood vessel set respectively.
The primitive vessel collection obtained when binary-state threshold is 14% in the present embodiment is as shown in figure 4, the corresponding obtained overall situation
The schematic diagram of blood vessel collection is as shown in Figure 5.As can be seen that it can effectively eliminate the ring as caused by taking pictures by removing segmentation blood vessel by mistake
Interference caused by the reasons such as shape is reflective, jump rank edge, plaque-like lesion and the bleeding lesion of non-vascular around optic disk, improves blood
The accuracy of pipe segmentation.
Position to obtain optic disk location information, idiographic flow such as Fig. 6 institutes by carrying out optic disk to eye fundus image in the present embodiment
Show, proceeded as follows for each global blood vessel collection：
(1-2) is directed to current global blood vessel and concentrates each blood vessel, and the convergence of the blood vessel is obtained using fuzzy convergence algorithm
Region；
Ballot of the number of convergence region belonging to (1-3) statistics each pixel of eye fundus image as the pixel
Value, and a ballot matrix is built according to the ballot value of each pixel, mean filter is carried out to ballot matrix, during mean filter
The size of the mean filter used is 6 × 6.
The pixel in each element and eye fundus image in the ballot matrix built in the present embodiment corresponds, to be right
The ballot value for the pixel answered.
(1-4) chooses the big preceding n pixel (n=3000 in the present embodiment) of ballot value according to matrix of being voted after filtering,
The regional connectivity algorithm based on eight connections is used to obtain several connected regions n pixel of selection, with each global blood
The maximum connected region of area corresponding to pipe collection makees the final convergence region of the global blood vessel collection, judges whether at least l
The overlapping region of final convergence region, wherein l=k/2, k are the number of default binary-state threshold, i.e. l=3：
If in the presence of the centre coordinate of the overlapping region maximum using area is used as optic disk location information；
Otherwise, optic disk location information is obtained with use specific template matching method.
(2) main blood vessel is determined according to the global blood vessel collection of binary-state threshold maximum (i.e. most loose) and optic disk location information,
And main blood vessel is classified to obtain main blood vessel classification information；
Main blood vessel is determined in the present embodiment by the following method：
It is neighbouring with the optic disk of determination using the region within the pixel of 100, optic disk center as optic disk adjacent domain
Length is more than the blood vessel conduct of default classification length threshold (default classification length threshold is 60 in the present embodiment) in region
Main blood vessel.
Following steps are classified to obtain main blood vessel classification information to main blood vessel in the present embodiment：
(2-1) obtains the average caliber of each main blood vessel, and it is vein blood vessel to specify the maximum main blood vessel of average caliber；
Each main blood vessel is cut into some fragments by (2-2), obtains corresponding main Vascular Slice；
Along the center line of main blood vessel when splitting in the present embodiment, each pixel is a section, and then will be each
Individual main blood vessel cutting (i.e. Vascular Slice) is some fragments.
The blood vessel center of (2-3) along the main Vascular Slice obtains the colouring information of 5 pixels to both sides respectively, and with institute
There is characteristic vector of the average of the colouring information of pixel as the main Vascular Slice；
The rgb value of the pixel and HSL values in the present embodiment, that is, the characteristic vector obtained are 6 dimensional vectors, and each dimension is right respectively
Should color value of the pixel on R, G, B and H, S, L * channel.
Then, feature based vector uses K mean cluster method to gather all main Vascular Slices for two classes, and with by venous blood
Class corresponding to pipe where main Vascular Slice is another kind of to be used as arterial blood tubing as venous blood tubing.
(2-4) is directed to each main blood vessel, and the classification results of the main blood vessel are used as using the class where more main Vascular Slice.
Such as any one main blood vessel, there is A% in arterial blood tubing in main Vascular Slice corresponding to it, B% exists
In venous blood tubing, if A is more than B, then it is assumed that the blood vessel is arteries, thinks that the blood vessel is vein blood vessel if A is less than B,
Otherwise, it is arbitrarily designated.
(3) blood vessel concentrated based on SAT breadth first search method to global blood vessel is used to enter using main blood vessel classification information
Row classification obtains global classification information.
Breadth first search is carried out to global blood vessel collection, three treaties are used in search procedure based on SAT breadth first search method
Beam condition carries out the transmission of blood vessel classification：Constraints 1：Two blood vessels of right-angled intersection are respectively labeled as two class blood vessels；Constraint
Condition 2：Three trouble structures in three vascular markers be；Constraints 3：Three blood vessels wherein blood vessel in three trouble structures is such as
When fruit and remaining two blood vessel angles sum are less than or equal to 270 degree, be determined as three trouble structures of constraints 2, i.e., three
Blood vessel is same class blood vessel, does not otherwise do and judges.
Fig. 7 is the global classification information obtained in the present embodiment using the breadth first search method based on SAT, can be with unfiled
The blood vessel number of (classify after still without classification information) is less, significantly smaller three to branch off in structures caused by the blood vessel segmentation
Blood vessel classification situation about can not carry out.
Do not make specified otherwise, Rounded Box represents obtained result in all flow charts in the present embodiment, and corner rectangle represents
Operation.
Technical scheme and beneficial effect are described in detail above-described embodiment, Ying Li
Solution is to the foregoing is only presently most preferred embodiment of the invention, is not intended to limit the invention, all principle models in the present invention
Interior done any modification, supplement and equivalent substitution etc. are enclosed, should be included in the scope of the protection.
Claims (6)
1. the arteriovenous retinal vessel sorting technique of a kind of eye fundus image, it is characterised in that comprise the following steps：
(1) eye fundus image is directed to, several different global blood vessel collection are obtained based on blood vessel segmentation method；
(2) for each global blood vessel collection, based on fuzzy convergence algorithm obtain corresponding to final convergence region, and according to it is each most
The overlapping cases of whole convergence region determine the optic disk center of the eye fundus image；
(3) a global blood vessel collection is selected, main blood vessel is determined according to the global blood vessel collection and optic disk location information, and to main blood vessel
Classified to obtain main blood vessel classification information；
(4) utilize described main blood vessel classification information to use to concentrate described global blood vessel based on SAT breadth first search method
Blood vessel classified to obtain global classification information；
When the step (1) obtains several different global blood vessel collection as follows based on blood vessel segmentation method, set first
Several fixed different binary-state thresholds, are then proceeded as follows for each binary-state threshold：
(1-1) carries out binary conversion treatment according to current binary-state threshold to described eye fundus image, and after extracting binary conversion treatment
Eye fundus image in center line and edge, obtain vascular tree；
(1-2) does disconnection process to described vascular tree crotch and obtains vessel segment, and enters line to each vessel segment and split
To blood vessel, primitive vessel collection is obtained；
(1-3) determines segmentation blood vessel by mistake, and concentrates to remove from primitive vessel and obtain global blood vessel collection；
Mistake segmentation blood vessel in the step (1-3) includes being divided into two classes, and the first kind splits the back of the body of the blood vessel based on blood vessel both sides by mistake
Scape difference determines that the second class is determined based on shape of blood vessel；
Proceeded as follows when convergence region corresponding to current global blood vessel collection is obtained in the step (2)：
(2-1) concentrates each blood vessel to current global blood vessel, and the convergence region of the blood vessel is obtained using fuzzy convergence algorithm；
Ballot value of the number of convergence region corresponding to (2-2) statistics each pixel of eye fundus image as the pixel；
(2-3) chooses several big preceding pixels of ballot value, and the area based on eight connections is used to several pixels of selection
Domain connection algorithm obtains several connected regions, and is used as final convergence region using the maximum connected region of area；
The step (2) determines regarding for the eye fundus image according to the overlapping cases of each final convergence region by the following method
Disk center：
The overlapping region of at least l final convergence regions is judged whether, wherein l=k/3~k/2, K are binary-state threshold
Number：
If in the presence of the centre coordinate of the overlapping region maximum using area is used as optic disk location information；
Otherwise, optic disk location information is obtained with use specific template matching method；
Carry out the transmission of blood vessel classification using three constraintss in search procedure based on SAT breadth first search method：Constraint
Condition 1：Two blood vessels of right-angled intersection are respectively labeled as two class blood vessels；Constraints 2：Three vascular markers in three trouble structures
For；Constraints 3：If three blood vessels wherein blood vessel in three trouble structures is less than with remaining two blood vessels angle sum
Or during equal to 270 degree, it is determined as three trouble structures of constraints 2, i.e., three blood vessels are same class blood vessel, otherwise do not do and judge.
2. the arteriovenous retinal vessel sorting technique of eye fundus image as claimed in claim 1, it is characterised in that the step
(3) main blood vessel is determined by the following method：
To be the region within several pixels as optic disk adjacent domain apart from optic disk center, with described optic disk proximity
Length is more than the blood vessel of default classification length threshold as main blood vessel in domain.
3. the arteriovenous retinal vessel sorting technique of eye fundus image as claimed in claim 2, it is characterised in that described point
Class length threshold is 50~65.
4. the arteriovenous retinal vessel sorting technique of eye fundus image as claimed in claim 3, it is characterised in that the step
(3) main blood vessel is classified as follows to obtain main blood vessel classification information：
(3-1) obtains the average caliber of each main blood vessel, and it is vein blood vessel to specify the maximum main blood vessel of average caliber；
Each main blood vessel is cut into some fragments by (3-2), obtains corresponding main Vascular Slice；
(3-3) extracts the characteristic vector of each main Vascular Slice, and will be described using clustering procedure based on described characteristic vector
Main Vascular Slice is gathered for two classes, and using the class where by main Vascular Slice corresponding to vein blood vessel as vein blood vessel, another kind of
As arteries；
(3-4) is directed to each main blood vessel, and the classification results of the main blood vessel are used as using the class where more main Vascular Slice.
5. the arteriovenous retinal vessel sorting technique of eye fundus image as claimed in claim 4, it is characterised in that the step
The characteristic vector of each main Vascular Slice is extracted in (3-3) by the following method：
The colouring information of all pixels point in the region within the blood vessel center of main Vascular Slice several pixels is obtained,
And the characteristic vector of the main Vascular Slice is used as using the average of the colouring information of all pixels point in the region.
6. the arteriovenous retinal vessel sorting technique of eye fundus image as claimed in claim 5, it is characterised in that extraction is each
The institute in the region within 5~8 pixels of blood vessel center of main Vascular Slice is obtained during the characteristic vector of main Vascular Slice
There is the colouring information of pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410850207.8A CN104573712B (en) | 2014-12-31 | 2014-12-31 | Arteriovenous retinal vessel sorting technique based on eye fundus image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410850207.8A CN104573712B (en) | 2014-12-31 | 2014-12-31 | Arteriovenous retinal vessel sorting technique based on eye fundus image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104573712A CN104573712A (en) | 2015-04-29 |
| CN104573712B true CN104573712B (en) | 2018-01-16 |
Family
ID=53089732
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410850207.8A Active CN104573712B (en) | 2014-12-31 | 2014-12-31 | Arteriovenous retinal vessel sorting technique based on eye fundus image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104573712B (en) |
Families Citing this family (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104809480B (en) * | 2015-05-21 | 2018-06-19 | 中南大学 | A kind of eye fundus image Segmentation Method of Retinal Blood Vessels based on post-class processing and AdaBoost |
| CN106204555B (en) * | 2016-06-30 | 2019-08-16 | 天津工业大学 | A kind of optic disk localization method of combination Gbvs model and phase equalization |
| CN106529420B (en) * | 2016-10-20 | 2019-07-19 | 天津大学 | The optic disk center positioning method of comprehensive eye fundus image marginal information and luminance information |
| CN106846301B (en) * | 2016-12-29 | 2020-06-23 | 北京理工大学 | Retina image classification method and device |
| CN106991718B (en) * | 2017-03-31 | 2022-02-15 | 上海健康医学院 | Method for reconstructing three-dimensional structure of fundus based on shading recovery |
| CN107203758A (en) * | 2017-06-06 | 2017-09-26 | 哈尔滨理工大学 | Diabetes patient's retinal vascular images dividing method |
| CN107229937A (en) * | 2017-06-13 | 2017-10-03 | 瑞达昇科技(大连)有限公司 | A kind of retinal vessel sorting technique and device |
| CN108182680B (en) * | 2017-12-28 | 2021-12-28 | 中科微光医疗研究中心(西安)有限公司 | IVOCT image-based angle automatic identification method for bifurcated vessels |
| CN108073918B (en) * | 2018-01-26 | 2022-04-29 | 浙江大学 | Method for extracting blood vessel arteriovenous cross compression characteristics of fundus retina |
| CN108230322B (en) * | 2018-01-28 | 2021-11-09 | 浙江大学 | Eye ground characteristic detection device based on weak sample mark |
| CN110276763B (en) * | 2018-03-15 | 2021-05-11 | 中南大学 | Retina blood vessel segmentation map generation method based on credibility and deep learning |
| CN108764286B (en) * | 2018-04-24 | 2022-04-19 | 电子科技大学 | Classification and identification method of feature points in blood vessel image based on transfer learning |
| CN108803994B (en) * | 2018-06-14 | 2022-10-14 | 四川和生视界医药技术开发有限公司 | Retinal blood vessel management method and retinal blood vessel management device |
| 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 |
| CN111696089B (en) * | 2020-06-05 | 2023-06-16 | 上海联影医疗科技股份有限公司 | Arteriovenous determination method, device, equipment and storage medium |
| CN111932554B (en) * | 2020-07-31 | 2024-03-22 | 青岛海信医疗设备股份有限公司 | Lung vessel segmentation method, equipment and storage medium |
| CN112734828B (en) * | 2021-01-28 | 2023-02-24 | 依未科技(北京)有限公司 | Method, device, medium and equipment for determining center line of fundus blood vessel |
| CN112734785B (en) * | 2021-01-28 | 2024-06-07 | 依未科技(北京)有限公司 | Method, device, medium and equipment for determining sub-pixel level fundus blood vessel boundary |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101393644A (en) * | 2008-08-15 | 2009-03-25 | 华中科技大学 | Hepatic portal vein tree modeling method and system thereof |
| CN102014731A (en) * | 2008-04-08 | 2011-04-13 | 新加坡国立大学 | Retinal image analysis systems and methods |
| CN102346911A (en) * | 2010-07-28 | 2012-02-08 | 北京集翔多维信息技术有限公司 | Method for segmenting blood vessel in digital subtraction angiography (DSA) image sequence |
-
2014
- 2014-12-31 CN CN201410850207.8A patent/CN104573712B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102014731A (en) * | 2008-04-08 | 2011-04-13 | 新加坡国立大学 | Retinal image analysis systems and methods |
| CN101393644A (en) * | 2008-08-15 | 2009-03-25 | 华中科技大学 | Hepatic portal vein tree modeling method and system thereof |
| CN102346911A (en) * | 2010-07-28 | 2012-02-08 | 北京集翔多维信息技术有限公司 | Method for segmenting blood vessel in digital subtraction angiography (DSA) image sequence |
Non-Patent Citations (2)
| Title |
|---|
| Blood vessel classification into arteries and veins in retinal images;Claudia Nieuwenhuis等;《Proceedings of SPIE - The International Society for Optical Engineering》;20070331;第1-8页，附图5和附图8 * |
| Locating the Optic Nerve in a Retinal Image Using the Fuzzy Convergence of the Blood Vessels;Adam Hoover等;《IEEE TRANSACTIONS ON MEDICAL IMAGING》;20030831;第22卷(第8期);第952-956页，附图6、9 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104573712A (en) | 2015-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104573712B (en) | Arteriovenous retinal vessel sorting technique based on eye fundus image | |
| CN104537669B (en) | The arteriovenous Segmentation Method of Retinal Blood Vessels of eye fundus image | |
| CN104545792B (en) | The arteriovenous retinal vessel optic disc localization method of eye fundus image | |
| CN104573716A (en) | Eye fundus image arteriovenous retinal blood vessel classification method based on breadth first-search algorithm | |
| AU2019205013B2 (en) | Method and system for identification of cerebrovascular abnormalities | |
| CN114926477B (en) | Brain tumor multi-mode MRI image segmentation method based on deep learning | |
| CN111046835A (en) | Eyeground illumination multiple disease detection system based on regional feature set neural network | |
| Seoud et al. | Automatic grading of diabetic retinopathy on a public database | |
| Muramatsu et al. | Automated detection and classification of major retinal vessels for determination of diameter ratio of arteries and veins | |
| CN108764342B (en) | Semantic segmentation method for optic discs and optic cups in fundus image | |
| CN109636805A (en) | A kind of uterine neck image lesion region segmenting device and method based on classification priori | |
| SG172159A1 (en) | A method and system for determining the position of an optic cup boundary | |
| CN109919915A (en) | Retina fundus image abnormal region detection method and device based on deep learning | |
| CN111797900B (en) | Artery and vein classification method and device for OCT-A image | |
| CN107203758A (en) | Diabetes patient's retinal vascular images dividing method | |
| CN110136151A (en) | The system and method for liver volume is obtained based on CT image | |
| Joshi et al. | Identification and reconnection of interrupted vessels in retinal vessel segmentation | |
| CN110246126A (en) | A method of extracting terminal bronchi tree from lung CT image | |
| Zhou et al. | Computer aided diagnosis for diabetic retinopathy based on fundus image | |
| CN116152492A (en) | Medical image segmentation method based on multi-attention fusion | |
| Yang et al. | Detection of microaneurysms and hemorrhages based on improved Hessian matrix | |
| CN111210436B (en) | Lens segmentation method, device and storage medium | |
| Jung et al. | Evaluating a deep-learning system for automatically calculating the stroke ASPECT score | |
| Zhang et al. | Application of the morphological ultimate opening to the detection of microaneurysms on eye fundus images from a clinical database | |
| Sreejini et al. | A review of computer aided detection of anatomical structures and lesions of DR from color retina images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20191008 Address after: 310000 room 410-2, floor 4, building 3, No. 9, Jiuhuan Road, Jianggan District, Hangzhou City, Zhejiang Province Patentee after: Hangzhou Qiushi innovative health technology Co., Ltd. Address before: 310027 Hangzhou, Zhejiang Province, Xihu District, Zhejiang Road, No. 38, No. Patentee before: Zhejiang University |
|
| 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: Classification of arteriovenous retinal vessels based on fundus images Effective date of registration: 20210423 Granted publication date: 20180116 Pledgee: Xixi sub branch of Bank of Hangzhou Co.,Ltd. Pledgor: Hangzhou Qiushi innovative health technology Co.,Ltd. Registration number: Y2021330000329 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20211216 Granted publication date: 20180116 Pledgee: Xixi sub branch of Bank of Hangzhou Co.,Ltd. Pledgor: Hangzhou Qiushi innovative health technology Co.,Ltd. Registration number: Y2021330000329 |