CN103854284B - Based on graphics search serous pigmentary epithelial pull-up from retina dividing method - Google Patents
Based on graphics search serous pigmentary epithelial pull-up from retina dividing method Download PDFInfo
- Publication number
- CN103854284B CN103854284B CN201410066419.7A CN201410066419A CN103854284B CN 103854284 B CN103854284 B CN 103854284B CN 201410066419 A CN201410066419 A CN 201410066419A CN 103854284 B CN103854284 B CN 103854284B
- Authority
- CN
- China
- Prior art keywords
- interface
- layer
- retina
- segmentation
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Eye Examination Apparatus (AREA)
Abstract
The invention discloses a kind of based on graphics search serous pigmentary epithelial pull-up from retina dividing method, including: (1) based on retina interface segmentation block speed B-scan image alignment method；(2) by interface significance degree order, dividing method is searched splitting the interface multi-resolution images as constraints；(3) carry out figure search algorithm by different constraints and obtain having interfacial method bottom the retina that under the pigment epithelial layer of rising zone, interface peace is sliding；(4) based on interface location difference bottom interface under pigment epithelial layer and retina, and the serous pigmentary epithelial pull-up of calmodulin binding domain CaM size and monochrome information from dividing method；(5) method carrying out outer retina Hierarchical Segmentation and correction after being planarized by image.Segmentation result of the present invention has higher accuracy, it is possible to substituting manual segmentation, the Clinics and Practices for clinically relevant ophthalmology disease can play important booster action.
Description
Technical field
The invention belongs to retinal image segmentation method, especially SD-OCT (domain optical coherence fault imaging) is regarded
Organisational level in nethike embrane image and the dividing method of lesion region.
Background technology
Retina is the extension of brain neuroblastoma tissue, has the multi-level institutional framework of complexity.SD-OCT technology has become
For the strong instrument of one of nondestructive evaluation retinal disease, in it is provided that quick, high-resolution, display retina
The partly 3-D view of layer, provides help for Clinical Ophthalmology doctor to diagnosis and the treatment of disease.Dividing of retina OCT image
Cut clinical practice significant: the segmentation of lesion region and disease is examined by its shape, size, the quantitative analysis of position
Disconnected and treatment has key effect；The segmentation of retinal tissue level and to various organization form, brightness quantitative analysis for sending out
Existing early lesion, the observation course of disease and research pathology all play an important role.But major part oculist uses manual side at present
The PVR that OCT is shown by formula carries out quantitative analysis, and subjectivity is strong, it is impossible to ensures accuracy and uniformity, and is difficult to
Analyzing three-dimensional scans the mass data brought comprehensively.
There is following defect in current retina OCT image automatic segmentation algorithm: (1) major part algorithm is all that two dimension is calculated
Method, i.e. independently splits in each sectioning image (x-z-plane image, referred to as B-scan image), and this kind of method is the most fully
Utilize three-dimensional contextual information, be easier to be affected by picture noise or artifact, cause segmentation errors.(2) major part is
Some retinal tissue hierarchical segmentation algorithm are both for normal retina design, produce relatively due to pathology when retinal tissue
During big deformation, these algorithms will lose efficacy.
Serous pigmentary epithelial pull-up is from being caused, such as age-related macular by multiple choroid/retinal disease
Sex change, polypoidal choroidal vasculopathy in Chinese patients, central serous chorioretinopathy, uveitis etc..So far, also
Not for serous pigmentary epithelial pull-up from retina OCT image in all distinguishable organisational levels and lesion region
The relevant report of the three-dimensional automatic division method of system.
Summary of the invention
The invention provides a kind of scheme solving the problems referred to above, provide first and a kind of there is feasibility and validity
For serous pigmentary epithelial pull-up from retina OCT image in all distinguishable organisational levels and the system of lesion region
Three-dimensional automatic division method.Wherein organisational level includes 10 layers: nerve fibre layer (NFL), ganglion-cell layer (GCL), interior
Plexiform layers (IPL), inner nuclear layer (INL), external plexiform layer (OPL), outer nuclear layer (ONL)+interior ganglionic layer (ISL), connection cilium (CL), outward
Ganglionic layer (OSL), verhoeff's membrane (VM), pigment epithelial layer (RPE), have 11 each interfaces.Add due to pigment epithelial layer
Departing from bottom retina, form a single interface bottom retina, therefore the present invention can detect 12 interfaces altogether.
The invention provides a kind of based on graphics search serous pigmentary epithelial pull-up from retina dividing method, should
Method mainly includes 5 steps:
Step S01, Image semantic classification: be substantially carried out the alignment between OCT denoising and B-scan image；
Step S02, the segmentation of each level of inner retina: use multi-resolution images search algorithm, according to interface contrast
Order from high to low is split successively, obtains nerve fibre layer (NFL), ganglion-cell layer (GCL), inner molecular layer (IPL), interior
Stratum nucleare (INL), external plexiform layer (OPL), the interface of outer nuclear layer (ONL)+interior ganglionic layer (ISL)；
Step S03, estimates bottom pigment epithelial layer segmentation and retina: in outer layer retinal area, with different constraints
Condition carries out figure search algorithm and obtains having interface bottom the retina that under the pigment epithelial layer of rising zone, interface peace is sliding；
Step S04, detachment of pigment epithelium region segmentation: demarcate bottom the retina that under pigment epithelial layer, interface peace is sliding
Region between face is detachment of pigment epithelium region, and removes flase drop region according to area size or monochrome information；
Step S05, the segmentation of each level of outer retina: with figure after image being planarized according to interface under pigment epithelial layer
The search algorithm detection each level of outer retina, obtains connecting cilium (CL), on outer ganglionic layer (OSL), verhoeff's membrane (VM), pigment
Interface between cortex (RPE).
Above-mentioned 5 steps are described in detail below,
(1) Image semantic classification
Image semantic classification mainly includes following two step: denoising and B-scan image alignment.
The denoising of (a) OCT image
The 3-D view that OCT ocular imaging instrument obtains contains more speckle noise.For ensureing the effect of subsequent singulation, must
The marginal information in image must be retained as far as possible while effectively removing noise.The present invention uses a kind of quickly two-sided filter
Each B-scan image is carried out denoising.Bilateral filtering result is:
Wherein
Here p is currently processed pixel, and q is the pixel in the neighborhood S of p, IpAnd IqIt is respectively the gray scale of p and q
Value,For the gray value of filter result,For normalization coefficient,WithIt is that standard deviation is respectively σsAnd σrGaussian function
Number, σsAnd σrThe two parameter carries out value according to picture size size and contrast on border size respectively.
(b) B-scan image alignment
During in imaging process, the motion of eye can cause continuous B-scan image, retinal location fluctuates up and down, i.e. image exists
Slice direction (y-direction) is upper discontinuous.This can cause difficulty to three-dimensional segmentation.The present invention divides based on to interface on retina
Cut result and carry out B-scan image alignment, because this interface contrast in all levels interface is the highest, even if in dislocation
Also can correctly split on image.Process with the segmentation of multi-resolution images search algorithm is as follows: to the 3-D view after denoising vertically
Carrying out down-sampling on direction (z-direction) makes direction pixel number become half, is repeated once this process, obtains three not
With the image of resolution ratio, it is expressed as yardstick 1,2,3 from low to high by resolution ratio.Segmentation is first on the yardstick 1 of lowest resolution
Carry out, on the basis of acquired results, yardstick 2 carries out near zone further Accurate Segmentation, the like, finally
Obtain the segmentation result on original image.Cutting procedure is the process of the divisional plane finding Least-cost, figure search algorithm complete.God
On fibrage, the cost function Sobel Operator at interface is calculated, by secretly to bright marginal position cost function relatively
Little.Distinguishing to i.e. be connected interface on cilium with ectonexine interface, on yardstick 1, cost function adds another point
Amount, this component is the brightness sum of some pixels above each picture point.So, due on nerve fibre layer interface be arranged above relatively
Dark background area, the cost of its correspondence is less than connecting the cost of interface location on cilium, it is possible to be correctly detected.Neural fine
After interface segments on dimension layer, every B-scan image calculates its average height, i.e. average z value.Get rid of during calculating
The point in image centre position, because these are affected by central fovea or pathology bigger displacement.Obtain according in every B-scan
To nerve fibre layer on move or move down this image in the average height of interface so that interface mean height on nerve fibre layer in result
Degree is a constant, just serves the effect of each image that aligns
(2) segmentation of each level of inner retina
Inner retina is affected less by pathology, the most first splits.Its dividing method is similar with step (1)
Multi-resolution images search method.First detection connects on cilium the interface at interface, i.e. ectonexine as constraints.It is segmented in
On nerve fibre layer, subgraph below interface is carried out, owing to pathology causes hydrops in retina, epiretinal part group
Being woven in OCT image and do not develop, the ectonexine interface therefore detected is actual for connecting interface and pigment epithelial layer on cilium
Lower surface merges, and it is defined as connect cilium pigment epithelial layer and merges interface.Then, according to each interface edge pair
Ratio degree order, with the interface that is partitioned into as constraints, segmentation ganglion-cell layer (GCL), inner molecular layer (IPL), inner nuclear layer
(INL), external plexiform layer (OPL), the interface of outer nuclear layer (ONL)+interior ganglionic layer (ISL).The interface possibility that boundary contrast is relatively low
Low yardstick cannot effectively be split, therefore need to start segmentation from higher resolution ratio.For eliminating the impact of noise, to obtain
Result carries out mean filter in the x direction, to obtain the segmentation interface smoothed.
(3) estimate bottom pigment epithelial layer segmentation and retina
Serous pigmentary epithelial pull-up from retina OCT image in, pigment epithelial layer is smooth in disengagement zone
Protuberance, the changing greatly in B-scan image front and back.And lower section is dark hydrops region, its bottom interface may not be developed.
The present invention, when based on figure search algorithm detection the two interface, uses identical cost function and different interface smoothness constraint
Condition.When smoothness constraint parameter is bigger, typically when smoothness constraint parameter takes 5～10, segmentation result is for there being local grand
Interface under the pigment epithelial layer risen.When parameter value is less, smoothness constraint parameter takes 1～4, time, segmentation result is smooth
Retina bottom interface.
(4) disengagement zone segmentation
Under step (3) pigment epithelial layer that obtains of segmentation, bottom interface and retina, region between interface is on pigment
Cortex sloughing off region.But owing to the impact of noise causes the error of segmentation local, interface, it is possible that the situation of flase drop.First
All pixels between interface bottom interface under pigment epithelial layer and retina are constituted several three-dimensional communication regions, point
Do not calculate volume and the mean flow rate in these regions.When volume is more than a certain predetermined value less than a certain predetermined value or mean flow rate
Time, it is believed that this region is that the disengagement zone of flase drop is removed.
While obtaining the disengagement zone of three-dimensional, it is also possible to obtain disengagement zone two-dimensional distribution in x-y direction.
This will be used in next step correction to outer retina segmentation result.
(5) segmentation of each level of outer retina
Outer retina is respectively organized in normal retina in flatter shape.When there is detachment of pigment epithelium,
These tissues swell also with the protuberance of pigment epithelial layer, and may not develop above rising zone, therefore scheme at OCT
In Xiang, these tissues are discontinuous, it is therefore necessary to plus certain constraints to ensure the correctness of segmentation.In the present invention, base
Under pigment epithelial layer, the result of interface segmentation, planarizes retinal images, will move up and down by each row in image so that look
Under element epithelial layer, interface becomes a plane.Thus pigment epithelial layer bump is reverted to smooth shape, also with regard to energy
Approximation recovers outer retina even shape under normal circumstances.On image after planarization, it is primarily based on step (2)
The connection cilium pigment epithelial layer that middle segmentation obtains merges interface, according to step (4) result, uses in detachment of pigment epithelium region
Second order polynomial carries out interpolation correction to obtain connecting interface on cilium.Then outer ganglionic layer (OSL), Wei Er are split with figure search algorithm
Conspicuous husband's film (VM), the interface of pigment epithelial layer (RPE), external ganglionic layer (OSL), the interface of verhoeff's membrane (VM), at pigment
Epithelial layer disengagement zone also needs to carry out interpolation correction by second order polynomial.By the connection cilium (CL) obtained in this step, acromere
Layer (OSL), verhoeff's membrane (VM), the interface of pigment epithelial layer (RPE) remap back and are finally split on original image
Result.
The present invention has merged bilateral filtering denoising, B-scan alignment, graphics cuts technology, connected region is split, image is smooth
The steps such as change, segmentation result correction, segmentation result has higher accuracy, it is possible to substitute manual segmentation, for clinically relevant
The Clinics and Practices of ophthalmology disease can play important booster action.
Accompanying drawing explanation
Fig. 1 is present configuration schematic diagram；
Fig. 2 is retinal tissue hierarchy chart picture, and Fig. 2 (a) is former B-scan image, and Fig. 2 (b) is 10 layers of normal retina
Tissue and 11 interfaces；
Fig. 3 is two-sided filter denoising result, and Fig. 3 (a) is original image, and Fig. 3 (b) is image after denoising；
Fig. 4 is B-scan alignment result x-y plane figure, Fig. 4 (a) original image, image after Fig. 4 (b) alignment；
Fig. 5 divides for connection cilium (CL), outer ganglionic layer (OSL), verhoeff's membrane (VM), the interface of pigment epithelial layer (RPE)
Cutting result, Fig. 5 (a) is interface segmentation result bottom interface and retina under pigment epithelial layer, and Fig. 5 (b) is according on pigment
Image after interface planarizes under cortex, Fig. 5 (c) is the segmentation result in Fig. 5 (b) to interface 7-10, and Fig. 5 (d) is for mapping
The result at the interface of cilium (CL), outer ganglionic layer (OSL), verhoeff's membrane (VM), pigment epithelial layer (RPE) is connected after returning original image
(wherein verhoeff's membrane (VM), pigment epithelial layer (RPE) interface substantially overlapping in this image)；
Fig. 6 is serous pigmentary epithelial detachment OCT image segmentation result, the two dimension display that Fig. 6 (a) is layering result, from
It is down 12 interfaces, Fig. 6 (b) is interface on nerve fibre layer, connect on cilium interface under interface, pigment epithelial layer,
The Three-dimensional Display of interface segmentation result bottom retina, Fig. 6 (c) is the two dimension display of disengagement zone segmentation, and Fig. 6 (d) is de-
Three-dimensional Display from region segmentation.
In Fig. 2, reference is as follows, 1 nerve fibre layer (NFL), 2 ganglion-cell layers (GCL), 3 inner molecular layers
(IPL), 4 inner nuclear layers (INL), 5 external plexiform layers (OPL), 6 outer nuclear layers (ONL)+interior ganglionic layer (ISL), 7 connect cilium (CL), outside 8
Ganglionic layer (OSL), 9 verhoeff's membranes (VM), 10 pigment epithelial layers (RPE).
Detailed description of the invention
Below in conjunction with detailed description of the invention, the present invention is expanded on further.
Shown in Figure 1, this method mainly includes 5 steps: the segmentation of each level of Image semantic classification, inner retina,
Estimate bottom pigment epithelial layer segmentation and retina, disengagement zone is split, the segmentation of each level of outer retina.
As in figure 2 it is shown, retinal tissue level includes 10 layers: nerve fibre layer (NFL) 1, ganglion-cell layer (GCL) 2,
Inner molecular layer (IPL) 3, inner nuclear layer (INL) 4, external plexiform layer (OPL) 5, outer nuclear layer (ONL)+interior ganglionic layer (ISL) 6, connection cilium
(CL) 7, outer ganglionic layer (OSL) 8, verhoeff's membrane (VM) 9, pigment epithelial layer (RPE) 10, have 11 interfaces.
The present invention a kind of based on graphics search serous pigmentary epithelial pull-up from retina dividing method, specifically describe
It is as follows,
(1) Image semantic classification
Image semantic classification mainly includes following two step: denoising and B-scan image alignment.
The denoising of (a) OCT image
The 3-D view that OCT ocular imaging instrument obtains contains more speckle noise.For ensureing the effect of subsequent singulation, must
The marginal information in image must be retained as far as possible while effectively removing noise.The present invention uses a kind of quickly two-sided filter
Each B-scan image is carried out denoising.Bilateral filtering result is:
Wherein
Here p is currently processed pixel, and q is the pixel in the neighborhood S of p.IpAnd IqIt is respectively the gray scale of p and q
Value.For the gray value of filter result,For normalization coefficient.WithIt is that standard deviation is respectively σsAnd σrGaussian function
Number.σsAnd σrThe two parameter carries out value according to picture size size and contrast on border size respectively.Denoising result such as Fig. 3
Shown in.
(b) B-scan image alignment
During in imaging process, the motion of eye can cause continuous B-scan image, retinal location fluctuates up and down, i.e. image exists
Slice direction (y-direction) is upper discontinuous.This can cause difficulty to three-dimensional segmentation.The present invention is based on to nerve fibre layer 1 upper bound
On face, i.e. retina, the segmentation result at interface carries out B-scan image alignment, because this interface is right in all levels interface
Higher than degree, even if also can correctly split on the image of dislocation.Nerve fibre layer 1 upper bound is split with multi-resolution images search algorithm
The process in face is as follows: the 3-D view after denoising carries out on vertical direction (z-direction) down-sampling and makes direction pixel
Number becomes half, is repeated once this process, obtains the image of three different resolutions, is expressed as yardstick by resolution ratio from low to high
1, yardstick 2 and yardstick 3.First segmentation is carried out, on the yardstick 1 of lowest resolution on the basis of acquired results, on yardstick 2
Further Accurate Segmentation is carried out near zone, the like, finally give the segmentation result on original image.
Cutting procedure is the process of the divisional plane finding Least-cost, figure search algorithm complete.Interface on nerve fibre layer 1
Cost function Sobel Operator be calculated, by secretly less to bright marginal position cost function.In order to ectonexine
Interface i.e. connects interface on cilium 7 and distinguishes, and on yardstick 1, cost function adds another component, and this component is each figure
The brightness sum of some pixels above picture point.So, owing on nerve fibre layer 1, interface is arranged above dark background area,
The cost of its correspondence is less than connecting the cost of interface location on cilium 7, it is possible to be correctly detected.Interface on nerve fibre layer 1
After having split, every B-scan image calculates its average height, i.e. average z value.Rejection image interposition during calculating
The point put, because these are affected by central fovea or pathology bigger displacement.Neural fine according to what every B-scan obtained
Move or move down this image in the average height of interface on dimension layer 1 so that in result, on nerve fibre layer 1, interface average height is one normal
Number, just serves the effect of each image that aligns.Image effect after alignment can be found out from x-y direction, as shown in Figure 4.
(2) segmentation of each level of inner retina
Inner retina is affected less by pathology, the most first splits.Its dividing method is similar with step (1)
Multi-resolution images search method.
First detection connects on cilium 7 interface at interface, i.e. ectonexine as constraints.It is segmented in nerve fibre layer
Carrying out in subgraph below interface on 1, owing to pathology causes hydrops in retina, epiretinal portion of tissue is schemed at OCT
Not developing in Xiang, the ectonexine interface therefore detected is actual for demarcating on interface and pigment epithelial layer 10 on connection cilium 7
Face merges, and it is defined as connect cilium pigment epithelial layer and merges interface.
Then, according to each interface contrast on border order, with the interface that is partitioned into as constraints, segmentation neuromere is thin
Born of the same parents' layer (GCL) 2, inner molecular layer (IPL) 3, inner nuclear layer (INL) 4, external plexiform layer (OPL) 5, outer nuclear layer (ONL)+interior ganglionic layer (ISL)
The interface of 6.The relatively low interface of boundary contrast cannot effectively may be split on low yardstick, therefore need to open from higher resolution ratio
Begin to split.Concrete segmentation order, upper and lower containment surfaces, the change of corresponding border and initial yardstick such as table 1 institute of multi-resolution segmentation
Show.For eliminating the impact of noise, the result obtained is carried out in the x direction mean filter, to obtain the segmentation interface smoothed.
(3) estimate bottom pigment epithelial layer segmentation and retina
Serous pigmentary epithelial pull-up from retina OCT image in, pigment epithelial layer is smooth in disengagement zone
Protuberance, the changing greatly in B-scan image front and back.And lower section is dark hydrops region, its bottom interface may not be developed.
The present invention, when based on figure search algorithm detection the two interface, uses identical cost function and different interface smoothness constraint
Condition.When smoothness constraint parameter takes 5～10, segmentation result is interface under the pigment epithelial layer having local eminence.When smooth
Degree constrained parameters take 1～4 constantly, and segmentation result is interface bottom smooth retina.
(4) disengagement zone segmentation
It is pigment that step (3) splits the region bottom the 10 times interfaces of pigment epithelial layer and retina obtained between interface
Epithelial layer disengagement zone.But owing to the impact of noise causes the error of segmentation local, interface, it is possible that the situation of flase drop.First
First all pixels between interface bottom 10 times interfaces of pigment epithelial layer and retina are constituted several three-dimensional communication districts
Territory, calculates volume and the mean flow rate in these regions respectively.When volume is more than a certain pre-less than a certain predetermined value or mean flow rate
During definite value, it is believed that this region is that the disengagement zone of flase drop is removed.
While obtaining the disengagement zone of three-dimensional, it is also possible to obtain disengagement zone two-dimensional distribution in x-y direction.
This will be used in next step correction to outer retina segmentation result.
(5) segmentation of each level of outer retina
Outer retina is respectively organized in normal retina in flatter shape.When there is detachment of pigment epithelium,
These tissues swell also with the protuberance of pigment epithelial layer, and may not develop above rising zone, therefore scheme at OCT
In Xiang, these tissues are discontinuous, it is therefore necessary to plus certain constraints to ensure the correctness of segmentation.
In the present invention, results based on 10 times interface segmentations of pigment epithelial layer, retinal images is planarized, will image
In each row move up and down so that 10 times interfaces of pigment epithelial layer become a plane.Thus by pigment epithelial layer bump
Revert to smooth shape, also just can approximate recovery outer retina even shape under normal circumstances.After planarization
Image on, be primarily based in step (2) the connection cilium pigment epithelial layer that segmentation obtains and merge interface,
According to step (4) result, carry out interpolation correction to be connected in detachment of pigment epithelium region second order polynomial
Connect interface on cilium 7.Then outer ganglionic layer (OSL) 8, verhoeff's membrane (VM) 9, pigment epithelial layer (RPE) are split with figure search algorithm
The interface of 10, segmentation order and constraints etc. are shown in Table 1 equally.To splitting outer ganglionic layer (OSL) 8, verhoeff's membrane (VM) 9
Interface, also needs in detachment of pigment epithelium region to carry out interpolation correction by second order polynomial.The connection that will obtain in this step
Cilium (CL) 7, outer ganglionic layer (OSL) 8, verhoeff's membrane (VM) 9, the interface of pigment epithelial layer (RPE) 10 remap back artwork
Final segmentation result is obtained on Xiang.
This step results is as shown in Figure 5.
Table 1 retina each level interface dividing method
(6) experimental result
Part of test results is as shown in Figure 6.
To Hierarchical Segmentation, with the mean value of two expert's separate manual segmentation results as goldstandard.Automatic segmentation result and
The absolute value of the difference of goldstandard segmentation interface z value is segmentation error.The absolute value of the difference of two expert's segmentation result z values represents to be seen
Difference between the person of examining.Experimental result on 20 samples shows, it is 2.25 ± 0.96 pixels (7.87 that the present invention splits mean error
± 3.36 microns), compared with difference between observer 2.23 ± 0.73 pixel (7.81 ± 2.56 microns), without obvious statistical discrepancy,
It is believed that it is essentially identical.Therefore this method can substitute manual segmentation method.
Pigment epithelial layer disengagement zone is split, using expert's manual segmentation result as goldstandard, uses true positives
Rate TPVF and false positive rate FPVF, as the objective indicator of appraisal procedure, are calculated as follows:
Wherein | | represent volume, CTPRepresent true positives point set, CFPRepresent false positive point set, CGTRepresent de-in goldstandard
From region point set, V represents all pixels set of whole retinal area.Test result indicate that, this method Average True positive rate
Being 87.9%, average false positive rate is 0.36%.
So far, a kind of be applicable to serous pigmentary epithelial pull-up from retina SD-OCT image automatic division method
Through realizing and being verified.The present invention has merged bilateral filtering denoising, B-scan alignment, graphics cuts technology, connected region is divided
Cut, image planarization, the step such as segmentation result correction, segmentation result has higher accuracy, it is possible to substitute manual segmentation, right
Clinics and Practices in clinically relevant ophthalmology disease can play important booster action.
The foregoing describe the general principle of the present invention, principal character and advantage.Skilled person will appreciate that of the industry, this
Invention is not restricted to the described embodiments, and the principle that the present invention is simply described described in above-described embodiment and specification, not
On the premise of departing from spirit and scope of the invention, the present invention also has various changes and modifications, and these changes and improvements both fall within
In scope of the claimed invention.Claimed scope is defined by appending claims and equivalent thereof.
Claims (7)
1. based on graphics search serous pigmentary epithelial pull-up from retina dividing method, it is characterised in that include following
Step,
Step S01, Image semantic classification: carry out the alignment between OCT image denoising and B-scan image；
Step S02, the segmentation of each level of inner retina: use multi-resolution images search algorithm, according to interface contrast from height
Split successively to low order, obtain nerve fibre layer, ganglion-cell layer, inner molecular layer, inner nuclear layer, external plexiform layer, outer core
The interface of layer+interior ganglionic layer；
Step S03, estimates bottom pigment epithelial layer segmentation and retina: in outer layer retinal area, by different smoothnesses about
Bundle parameter carries out figure search algorithm and obtains having interface bottom the retina that under the pigment epithelial layer of rising zone, interface peace is sliding；
Step S04, detachment of pigment epithelium region segmentation: district between interface bottom interface and retina under pigment epithelial layer
Territory is detachment of pigment epithelium region, removes flase drop region according to area size or monochrome information；
Step S05, the segmentation of each level of outer retina: search calculation with figure after being planarized by image according to interface under pigment epithelial layer
The method detection each level of outer retina, obtains connecting cilium, outer ganglionic layer, verhoeff's membrane, the interface of pigment epithelial layer；
Described step S04, detachment of pigment epithelium region segmentation specifically includes,
All pixels between interface bottom interface under pigment epithelial layer and retina are constituted several three-dimensional communication districts
Territory, calculates volume and the mean flow rate in described three-dimensional communication region respectively, when the volume in region is less than a certain predetermined value or average
When brightness is more than a certain predetermined value, described region is the disengagement zone of flase drop.
The most according to claim 1 based on graphics search serous pigmentary epithelial pull-up from retina dividing method,
It is characterized in that, in step S01, OCT image denoising specifically includes following steps, uses a kind of quickly two-sided filter to each B
Scan image obtains the 3-D view after denoising after carrying out denoising, bilateral filtering result such as formula (1a):
Wherein
P is currently processed pixel, and q is the pixel in the neighborhood S of p, IpAnd IqIt is respectively the gray value of p and q,For filter
The gray value of ripple result,For normalization coefficient,WithIt is that standard deviation is respectively σsAnd σrGaussian function, σsAnd σrThis
Two parameters carry out value according to picture size size and contrast on border size respectively.
The most according to claim 1 based on graphics search serous pigmentary epithelial pull-up from retina dividing method,
It is characterized in that, being aligned between described B-scan image carries out B-scan figure based on to the segmentation result at interface on nerve fibre layer
As alignment, specifically include following steps,
1-1) with interface on multi-resolution images search algorithm segmentation nerve fibre layer: to the 3-D view in the vertical direction after denoising
Carrying out down-sampling makes direction pixel number become half, is repeated once this process, obtains the image of three different resolutions,
Yardstick 1, yardstick 2, yardstick 3 it is expressed as from low to high by resolution ratio；
1-2) split on the yardstick 1 of lowest resolution, on the basis of yardstick 1 splits acquired results, attached on yardstick 2
Carrying out further Accurate Segmentation near field, then on yardstick 3, near zone continues segmentation, finally gives on original image
Segmentation result；
After 1-3) interface segments on nerve fibre layer, every B-scan image calculates its average height, i.e. averagely z value,
The point in rejection image centre position during calculating, according to interface average height on the nerve fibre layer obtained in every B-scan
Upper shifting or move down this image so that in result, on nerve fibre layer, interface average height is constant, just serves each image of alignment
Effect.
The most according to claim 3 based on graphics search serous pigmentary epithelial pull-up from retina dividing method,
It is characterized in that: step 1-2) described in be divided into the process of divisional plane finding Least-cost, figure search algorithm complete, neural
On fibrage, the cost function Sobel Operator at interface is calculated, by secretly less to bright marginal position cost function,
Distinguishing to i.e. be connected interface on cilium with ectonexine interface, on yardstick 1, cost function adds one-component, institute
State the brightness sum that component is some pixels above each picture point.
The most according to claim 1 based on graphics search serous pigmentary epithelial pull-up from retina dividing method,
It is characterized in that, the segmentation of each level of described inner retina specifically includes following steps,
First on detection connection cilium, the interface at interface, i.e. ectonexine, as constraints, is segmented in the nerve fibre layer upper bound
Subgraph below face is carried out, owing to pathology causes hydrops in retina, epiretinal portion of tissue in OCT image not
Development, forms ectonexine interface actual the merging for interface on interface on connection cilium and pigment epithelial layer therefore detected
, it is defined as connects cilium pigment epithelial layer and merges interface；
Then, according to above-mentioned each interface contrast on border order, with the interface that is partitioned into as constraints, segmentation neuromere is thin
Born of the same parents' layer, inner molecular layer, inner nuclear layer, external plexiform layer, outer nuclear layer+internal segment bed boundary.
The most according to claim 1 based on graphics search serous pigmentary epithelial pull-up from retina dividing method,
It is characterized in that, estimate to specifically include following steps bottom the segmentation of described step S03 pigment epithelial layer and retina, search based on figure
Under algorithm detection pigment epithelial layer bottom interface and retina during interface, identical cost function and different interfaces is used to put down
Smoothness constraints condition, when smoothness constraint parameter takes 5～10, segmentation result is the pigment epithelial layer lower bound having local eminence
Face, when smoothness constraint parameter takes 1～4, segmentation result is interface bottom smooth retina.
The most according to claim 1 based on graphics search serous pigmentary epithelial pull-up from retina dividing method,
It is characterized in that, the segmentation of the described each level of step S05 outer retina specifically includes following steps,
Based on the result of interface segmentation under pigment epithelial layer, retinal images is planarized, will image move up and down by each row,
Make interface under pigment epithelial layer become a plane, thus pigment epithelial layer bump reverted to smooth shape,
Recover outer retina even shape under normal circumstances；
On retinal images after planarization, merge based on step S02 is split the connection cilium pigment epithelial layer obtained
Interface, according to the result of step S04, in detachment of pigment epithelium region, second order polynomial carries out interpolation correction to be connected
Interface on cilium, splits outer ganglionic layer, verhoeff's membrane, PE bed boundary, external ganglionic layer, verhoeff's membrane with figure search algorithm
Interface, in detachment of pigment epithelium region, second order polynomial carries out interpolation correction, by connection cilium obtained above, acromere
Layer, verhoeff's membrane, PE bed boundary remap back and obtain final segmentation result on original image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410066419.7A CN103854284B (en) | 2014-02-26 | 2014-02-26 | Based on graphics search serous pigmentary epithelial pull-up from retina dividing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410066419.7A CN103854284B (en) | 2014-02-26 | 2014-02-26 | Based on graphics search serous pigmentary epithelial pull-up from retina dividing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103854284A CN103854284A (en) | 2014-06-11 |
| CN103854284B true CN103854284B (en) | 2016-08-31 |
Family
ID=50861896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410066419.7A Active CN103854284B (en) | 2014-02-26 | 2014-02-26 | Based on graphics search serous pigmentary epithelial pull-up from retina dividing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103854284B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106997594B (en) * | 2016-01-26 | 2020-12-01 | 上海联影医疗科技股份有限公司 | Method and device for positioning eye tissue |
| WO2017171643A1 (en) * | 2016-03-28 | 2017-10-05 | Agency For Science, Technology And Research | Three-dimensional representation of skin structure |
| JP2020513983A (en) | 2017-03-23 | 2020-05-21 | ドヘニー アイ インスティテュート | Multiplex en-face angiographic averaging system, method and apparatus for optical coherence tomography |
| CN109886969B (en) * | 2019-01-15 | 2023-03-17 | 南方医科大学 | Three-dimensional automatic segmentation method for respiratory tract endoscopic optical coherence tomography image |
| CN110010219B (en) * | 2019-03-13 | 2021-12-10 | 杭州电子科技大学 | Intelligent detection system and detection method for retinopathy by optical coherence tomography |
| CN110163838B (en) * | 2019-04-01 | 2023-04-18 | 江西比格威医疗科技有限公司 | Retina OCT image layering algorithm |
| CN117893742B (en) * | 2024-03-15 | 2024-05-28 | 深圳大学总医院 | Uveitis image feature recognition method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103514605A (en) * | 2013-10-11 | 2014-01-15 | 南京理工大学 | Choroid layer automatic partitioning method based on HD-OCT retina image |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5884195B2 (en) * | 2010-07-21 | 2016-03-15 | ディオプシス、インコーポレーテッド | Method for analyzing OCT image and method for analyzing optic nerve |
-
2014
- 2014-02-26 CN CN201410066419.7A patent/CN103854284B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103514605A (en) * | 2013-10-11 | 2014-01-15 | 南京理工大学 | Choroid layer automatic partitioning method based on HD-OCT retina image |
Non-Patent Citations (3)
| Title |
|---|
| Intraretinal Layer Segmentation of Macular Optical Coherence Tomography Images Using Optimal 3-D Graph Search;Mona K. Garvin et al.;《IEEE TRANSACTIONS ON MEDICAL IMAGING》;20081010;第27卷(第10期);图2，3，第1496页左栏，第1497页右栏.第1498页右栏，1500页右栏 * |
| Validated Automatic Segmentation of AMD Pathology Including Drusen and Geographic Atrophy in SD-OCT Images;Stephanie J. Chiu et al.;《Clinical Trials》;20120131;第53卷(第1期);第55页右栏，图4 * |
| 双边滤波算法的快速实现及其在图像处理的应用;李俊峰;《中国优秀硕士学位论文全文数据库-信息科技辑》;20131231;第2.5.1节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103854284A (en) | 2014-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103854284B (en) | Based on graphics search serous pigmentary epithelial pull-up from retina dividing method | |
| US7992999B2 (en) | Automated assessment of optic nerve head with spectral domain optical coherence tomography | |
| JP6025311B2 (en) | Ophthalmic diagnosis support apparatus and method | |
| Abràmoff et al. | Retinal imaging and image analysis | |
| KR101373935B1 (en) | Ophthalmic apparatus, control method of an ophthalmic apparatus and computer readable storage medium | |
| EP3568059B1 (en) | Method and system for three-dimensional thickness mapping of corneal micro-layers and corneal diagnoses | |
| US9098742B2 (en) | Image processing apparatus and image processing method | |
| CN102458222B (en) | Image processing apparatus and image processing method | |
| KR20130100384A (en) | Image processing apparatus, control method thereof, and computer program | |
| WO2010017356A2 (en) | Establishing compatibility between two-and three dimensional optical coherence tomography scans | |
| JP2013153844A (en) | Image processing apparatus, image processing method, and program | |
| Hassan et al. | Fully automated detection, grading and 3D modeling of maculopathy from OCT volumes | |
| de Sisternes et al. | Individual drusen segmentation and repeatability and reproducibility of their automated quantification in optical coherence tomography images | |
| JP2017127397A (en) | Image processing device, estimation method, system and program | |
| US10573007B2 (en) | Image processing apparatus, image processing method, and image processing program | |
| Mokhtari et al. | Local comparison of cup to disc ratio in right and left eyes based on fusion of color fundus images and OCT B-scans | |
| EP3417401B1 (en) | Method for reducing artifacts in oct using machine learning techniques | |
| Dufour et al. | Pathology hinting as the combination of automatic segmentation with a statistical shape model | |
| EP3129955B1 (en) | Method for the analysis of image data representing a three-dimensional volume of biological tissue | |
| de Sisternes et al. | A machine learning approach for device-independent automated segmentation of retinal cysts in spectral domain optical coherence tomography images | |
| WO2019157113A1 (en) | Segmentation-based corneal mapping | |
| Guimarães et al. | Three-dimensional segmentation and reconstruction of the retinal vasculature from spectral-domain optical coherence tomography | |
| WO2022175945A1 (en) | System and method for monitoring optic disc conditions | |
| CN115413350A (en) | Bruch's membrane segmentation in OCT | |
| Jo et al. | Comparison of evaluation parameters in the retinal layer between diabetic cystoid macular edema and postoperative cystoid macular edema after cataract surgery based on a hierarchical approach |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20170810 Address after: 215011 No. 209 Chuk Yuen Road, Suzhou hi tech Development Zone, Jiangsu, Suzhou Patentee after: Suzhou were Medical Technology Co. Ltd. Address before: 215006, 1, ten Zi street, Suzhou District, Jiangsu, Suzhou Patentee before: Soochow University |
|
| TR01 | Transfer of patent right |