DE102008018701A1 - Method for automated detection and segmentation of the papilla in fundus images - Google Patents

Method for automated detection and segmentation of the papilla in fundus images

Info

Publication number
DE102008018701A1
DE102008018701A1 DE102008018701A DE102008018701A DE102008018701A1 DE 102008018701 A1 DE102008018701 A1 DE 102008018701A1 DE 102008018701 A DE102008018701 A DE 102008018701A DE 102008018701 A DE102008018701 A DE 102008018701A DE 102008018701 A1 DE102008018701 A1 DE 102008018701A1
Authority
DE
Germany
Prior art keywords
papilla
starting point
method
fundus
step
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.)
Withdrawn
Application number
DE102008018701A
Other languages
German (de)
Inventor
Axel Dr. Doering
Torsten Schmidt
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Carl Zeiss Meditec AG
Original Assignee
Carl Zeiss Meditec AG
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Carl Zeiss Meditec AG filed Critical Carl Zeiss Meditec AG
Priority to DE102008018701A priority Critical patent/DE102008018701A1/en
Publication of DE102008018701A1 publication Critical patent/DE102008018701A1/en
Withdrawn legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/149Segmentation; Edge detection involving deformable models, e.g. active contour models
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/12Edge-based segmentation
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30004Biomedical image processing
    • G06T2207/30041Eye; Retina; Ophthalmic

Abstract

The present invention relates to a method by which the edge (disc) of the papilla can be detected in fundus recordings and whose form can be mathematically (approximately) reproduced. In the method according to the invention for the automated detection and segmentation of the papilla in fundus images based on fundus recordings in suitable spectral regions of a color photograph, with fully imaged papilla, four sectors are determined after the papilla region is located and the starting point Mn with n = 0 is selected to determine the gray value gradients in appropriate spectral ranges, which calculates gradient curves from the gray scale curves of the four sectors and determines from the global minima in each sector of the center of the resulting contour as starting point Mn + 1 for the new method step b). The procedure is completed when the deviation between the starting point Mn + 1 and the starting point Mn is within a specified tolerance. The method according to the invention for automated detection and segmentation of the papilla in fundus images is intended in particular for use in fundus cameras.

Description

  • The The present invention relates to a method by which the edge (disc) the papilla is detected in fundus recordings and its form is mathematical (Approximated) can be replicated.
  • Of the Entry point of the optic nerve into the fundus is called the papilla denotes, appears as a light, slightly elliptical Area and causes in the visual field the "blind spot". This entry point of the optic nerve is a polygonal, flattened Bulging of the conjunctiva, in the center of a fine vascular tree which is susceptible to allergic inflammation is.
  • The Information of the shape of the papilla forms the basis for an ophthalmological examination, especially of the green Stars (glaucoma). At the glaucoma examination represents the bulge the papilla is a well-studied, valid indicator of the onset and progression of the disease. Glaucoma represents irreversible damage and at the same time the most common diseases of the optic nerve.
  • Of the Green Star is known as a disorder of vision. In the study of the green star, the observation applies of the blind spot of the patient on the basis of a picture of the patient Ocular fundus (fundus) as crucial. In recent years It has become possible, as a result of the blind spot Progress in the technology of recording evaluation quantitatively to measure and evaluate.
  • The Glaucoma is often diagnosed very late, because in the early stage of the disease the peripheral vision decreases and thus is not perceived by the patient. With progressive Disease course these visual field failures occur in Area of the macula.
  • Under the macula is an approx. 5 mm large area at the posterior pole the retina, below the papilla, the largest Visual acuity and also referred to as "yellow spot" becomes.
  • by virtue of the slow course of the disease in glaucoma is of the greatest Meaning, this disease by means of screening method in an early Recognize disease stage.
  • To The prior art discloses various methods with which the papilla detected in Fundusaufnahmen and their shape is modeled using a mathematical model.
  • So will in the DE 697 31 167 T2 a method and a device for the evaluation of stereo images of the ocular fundus described, with which a blind spot on the basis of stereo images of the fundus can be detected and evaluated. With the help of stereo recordings different height points (lowest and highest points) are determined from the three-dimensional data of the fundus and from this an outer peripheral line of the blind spot is determined. The papilla margin is characterized by a manual marking of several bases. A segmentation with an ellipse is not provided. A disadvantage of this solution has the effect that stereo recordings are required for the determination of an outer circumferential line of the blind spot, since otherwise no height points can be determined.
  • Of the in [1] of Xu and others described algorithm is also based on active contours, however, is for the intended Use too expensive, since the papilla with any Contour is approximated. A segmentation of the papilla with an ellipse is sufficient for a glaucoma examination. One Another disadvantage of the algorithm described by Xu and others is to be seen in the specification of an initial contour. This should be as close to the actual papilla margin lie and must be given manually or by another algorithm be calculated.
  • in the By contrast, Tang and others in [2] make an algorithm described, with which the papilla also approximated with an ellipse can be. By the mathematical methods used is the mathematical effort, however, significantly higher, which leads to higher computation times, so the procedure hardly suitable for a screening-capable fundus camera is. Moreover, this is analogous to the solution also assumed an initial contour according to [1].
  • The most significant disadvantage of the known in the prior art solutions for glaucoma detection is the fact that the disc and / or cup area of the papilla is manually segmented by adaptable mathematical forms, such as circles and ellipses preferably by the physician. On the one hand this is time consuming and on the other hand the accuracy of the segmentation depends on the experiences and the skill of the doctor.
  • One Another disadvantage is the fact that different methods based on stereo image data, as these are not always or only with considerable additional equipment are available.
  • Literature:
  • Of the The present invention is based on the object, a method for a screening-capable fundus camera for recording develop digital fundus images, with which an automated Measurement of the papilla in these fundus images possible is. The required computational effort should be so be kept as low as possible.
  • According to the invention the object by the features of the independent claims solved. Preferred developments and refinements are the subject of the dependent claims.
  • In the method according to the invention for the automated detection and segmentation of the papilla in fundus images, based on fundus recordings in suitable spectral ranges, for example in the green channel of a color photograph, with fully imaged papilla, after the papilla region is located and the starting point M n with n = 0 is selected four Sectors for determining the gray scale gradients in suitable spectral ranges, for example in the red channel of a color image, the gradient curves calculated from the gray scale gradients of the four sectors and from the global minima in each sector of the center of the resulting contour as the starting point M n + 1 for the process step b ) certainly. The method is completed when the deviation between the starting point M n + 1 and the starting point M n is within a specified tolerance.
  • When Prerequisite for this is an algorithm for the detection of To develop papilla and segmentation of the disc area and to try it out.
  • On Basis of an Overview of Algorithms for Contour tracking and segmentation required for optic disc identification and segmentation should be a procedure for the detection of the papilla as well as for the segmentation of the disc area be developed or compiled.
  • The Investigation of the efficiency of the process by means of a prototype in terms of sensitivity and specificity respectively. The procedure is intended for use in the fundus imaging system Carl Zeiss Meditec AG as a software library.
  • Even though the inventive method for automated Detection and segmentation of the papilla in fundus recordings in particular for use in screening-capable fundus cameras is intended, it can be used in principle for all fundus cameras, that have a digital image capture, be used.
  • A so-called fundus imaging system offers the possibility of electronic image acquisition and thus a direct diagnosis and examination documentation. Such a system is available with the VISUCAM PRO NM © and VISUPAC © from Carl Zeiss Meditec AG. In addition to the management of patient information, different findings can be performed and documented using the Graphical Findings Editor (GFE). With the nonmydriatic fundus camera VISUCAM PRO NM © , images can be taken without an extended (mydrialized) patient pupil with a resolution of 2196 × 1958 pixels at a viewing angle of 45 ° and images with 1620 × 1444 pixels at 30 °. The reproduction scale is 0.006 mm / pixel and is identical for both recording formats. The fundus images are stored as lossy, JPEG-compressed RGB color images with a color depth of 8-bits per color channel in a database and can serve as a basis for the inventive method for automated detection and segmentation of the papilla.
  • The procedure works extremely reliably, ie with a high success rate if the papilla in the Fundus images is completely displayed, the fundus images have a sufficiently high contrast and as possible no papilla-like artifacts or exudates (in the sense of: size, shape and color) included.
  • The Invention will be described below with reference to an embodiment described in more detail. Show this
  • 1 : the procedure in the form of a flowchart,
  • 2 a fundus image with the localized center M o and the fixed sectors 1 to 4,
  • 3a : the statically determined gray values of the gray value curve of sector 1 and
  • 3b : the gray values determined dynamically from sector 1
  • The inventive method for automated detection and segmentation of the papilla in fundus images, based on colored fundus images with fully imaged papilla and can be divided into the following process steps:
    • a) The area of the papillae is located and the starting point M n with n = 0 is selected.
    • b) Definition of four sectors to determine the gray value gradients.
    • c) Calculation of the gradients from the gray scale gradients of the four sectors.
    • d) determining the global minimum of an energy function over the set of all possible combinations of local minima of the gradient curves.
    • e) determining the center point of the resulting contour as the starting point M n + 1 for the renewed method step b).
    • f) The process is ended when the deviation between the starting point M n + 1 and the starting point M n is within a specified tolerance.
  • The inventive method is in 1 as a flowchart, starting from the localization of the pupil area and the selection of the starting point M n , up to the comparison of the coordinates of the last two starting points M n + 1 and M n and the completion of the segmentation shown.
  • Step a: Localization of the papilla area and definition of the starting point M n with n = 0.
  • In method step a), the area of the papilla is located and the starting point M o is selected. This is done by converting the colored fundus image into a binary image, detecting the binary objects and defining the center of gravity of the largest binary object as the starting point M o . This shows 2 a fundus image with the localized center M o and the specified sectors 1 to 4.
  • The Gray value conversion can be done with different methods respectively. Due to the pictures available in the RGB color space the fundus camera can use a color channel directly as a grayscale image become. In addition, it is possible the intensity channel I or the brightness channel L of the color space to use.
  • While the intensity channel is formed from the arithmetic mean of the three color channels of the RGB color space,
    Figure 00080001
    The brightness channel represents the average of the minimum values and the maximum values of the three color channels of the RGB color space.
    Figure 00080002
  • The arithmetic mean of these two values then forms the gray value, where
    Figure 00080003
    the intensities of the considered pixel f (x; y) and max (f RGB (x; y)) represent their maximum and min (f RGB (x; y)) their minimum.
  • After this gray value conversion or binarization of the color image is carried out under consideration of a Threshold classifies each pixel, separating the background and object pixels.
  • in the Following this, the individual object pixels become binary objects joined together, the neighborhood relationships of the individual object pixels are taken into account among each other. For a localization of the papilla, which is usually the largest coherent object in the fundus image accordingly, becomes the largest binary object searched.
  • The center of gravity of all object pixels in the largest binary object, which corresponds to the papilla, is evaluated as the papilla point and thus defined as the starting point M o .
  • An advantageous embodiment can be seen in that in step a) detected binary objects are compared with a typical papilla binary object and the center of gravity of the largest binary object is only defined as the starting point M o when the largest binary object corresponds to a typical papilla binary object.
  • A Another advantageous embodiment results from the fact that a normative database with average papilla parameters different ethnic groups can easily be involved in the process.
  • Step b: Definition of four Sectors for determining gray scale gradients.
  • Again 2 can be seen, the area around the starting point M o is divided into four sectors, the sectors 1 and 3 in horizontal and the sectors 2 and 4 in the vertical direction. Starting from the starting point M o , the gray-scale gradients f (u) are created on the search beams and formed by the mean value of the gray values on the associated circular arcs. It can be minimized by a large number of gray levels on the corresponding arc of the disturbing influence of the blood vessels.
  • The radii of the circular arcs can be defined statically or dynamically. While at the in 3a The radial radii of the circular arcs are constant and are determined by means of the statistical papilla parameters, the radii of the circular arcs at the in 3b shown dynamic variant of the distance from the current starting point M o .
  • Step c: Calculation of Gradient Gradients from the gray value gradients of the four sectors.
  • in the Process step c), the calculation of the gradient curves takes place from the gray value gradients of the four sectors.
  • There the papilla has higher gray values than the fundus, has the gradient at the edge of the papilla a minimum. By disturbances inside and outside the papilla There are several local minima per sector. That's why the determined local minima in each search direction the papilla margin or also represent variations within the papilla.
  • Becomes a minimum used in the further procedure, which is not the papilla margin represents, it can lead to faulty segmentation results come. For this reason, any combination of local minima each search direction considered as a possible solution become.
  • Process step d: Determination of the global Minimums of energy function over the set of all possible combinations local minima of the gradients.
  • in the Process step d), the determination of the energy function from the determined amounts of the gradients, the axial lengths and the ratio of the axial lengths of the resulting Contour for all combinations of minima of all search directions.
  • Analogous to active contours was used for the evaluation of each Combinations defined an energy function. By inclusion The determined statistical papilla parameter becomes the combination local minima of the gradients in each sector the lowest energy determined. Combinations with axial lengths or axis length ratios outside The statistical results are discarded and not considered further.
  • requirement for the definition of a static parameter set, that all terms of the energy function describe the same amount of energy. To achieve this, each term is above all combinations interpreted as a vector and normalized to the range [0 ... 1] (normalized Energy function). The positions of the minima of the combination with The lowest normalized total energy is the four vertices the resulting contour.
  • Method step e: Determining the center point of the resulting contour as the starting point M n + 1 for the renewed method step b).
  • In method step e), the determination of the center point of the resulting contour takes place as starting point M n + 1 for the renewed method step b).
  • There the combination of minima determined in process step d) the lowest total energy the four vertices of the resulting Forms the contour, the center of the contour opens up to determine a simple way.
  • This center also represents the starting point M n + 1 for the next iteration.
  • Method step f: The method is ended when the deviation between the starting point M n + 1 and the starting point M n is within a specified tolerance.
  • In this method step, the just-determined new starting point M n + 1 is compared with the starting point M n . If the distance of the starting points lies within a specified tolerance, the method is ended. Otherwise, the process starts again with method step b). In the method, a deviation of less than 3 pixels is considered suitable. However, the accuracy of the method can be significantly increased if the deviation is limited to a maximum of 1 pixel.
  • Active circular arc model
  • In a particularly advantageous embodiment of the invention Method is used as the resulting contour an ellipse.
  • Various Evaluations had shown that the contour of a papilla only in limited extent can be approximated with a circle. Essential more precisely, the segmentation is possible with an ellipse.
  • Out For this reason, the method according to the invention developed on the basis of an active circular arc model. This Method is based on the iterative fitting of the four vertices an ellipse to the edge area of the papilla.
  • Of the Papilla edge can thus by simple mathematical means using approximated to an ellipse, d. H. be segmented.
  • With the inventive method for automated Detection and segmentation of the papilla in fundus images, based on colored fundus images with fully mapped Papilla, a solution is provided with an automated measurement of the papilla in fundus images is possible without the required Computing effort is disproportionately large.
  • One particular advantage of the proposed method is to be seen in that the segmented papilla as a reference point for a Fundus coordinate system can be used.
  • From Another advantage is that the process takes place on stereo image data based on colored fundus images, with much higher Probability are available as stereo image data.
  • QUOTES INCLUDE IN THE DESCRIPTION
  • This list The documents listed by the applicant have been automated generated and is solely for better information recorded by the reader. The list is not part of the German Patent or utility model application. The DPMA takes over no liability for any errors or omissions.
  • Cited patent literature
    • - DE 69731167 T2 [0009]

Claims (9)

  1. Method for the automated detection and segmentation of the papilla in fundus images, based on fundus images with fully imaged papilla, with the following method steps: a) the papilla area is located and the starting point M n is selected with n = 0, b) four sectors are defined to determine the gray value gradients d) determining the global minimum of an energy function over the set of all possible combinations of local minima of the gradient curves e) determining the center point of the resulting contour as starting point M n + 1 for the gradient new process step b) f) the process is ended when the deviation between the starting point M n + 1 and the starting point M n is within a specified tolerance.
  2. The method of claim 1, wherein in step a) the area of the optic disc is located and the starting point M o is selected by converting the colored fundus image into a binary image, detecting the binary objects and defining the center of gravity of the largest binary object as the starting point M o .
  3. Method according to claims 1 and 2, wherein in step a) the binary objects are detected and compared with a typical papillary binary object and the center of gravity of the largest binary object is defined as starting point M o only if the largest binary object corresponds to a typical papillary binary object ,
  4. The method of claim 1, wherein in the step b) the gray value gradients in the four sectors are determined, by gray scale gradients in horizontal (sectors 1 and 3) and in the vertical direction (sectors 2 and 4) created and the individual gray values of each gray value curve from the mean value the gray values along the concentric arc with given, fixed radius of the corresponding sector become.
  5. Method according to Claims 1 and 4, in which, in method step b), the individual gray values of each gray value curve are determined from the mean of the gray values along the circular arc of the corresponding sector, the radii of the circular arcs being dependent on the distance to the current starting point M o .
  6. The method of claim 1, wherein in the step c) the calculation of the gradient curves from the gray value gradients The four sectors are made by taking the local minima in each search direction determined and their combination as a possible solution (ie as edge locations of a circumscribing the area of the optic disc Ellipse).
  7. The method of claim 1 and 6, wherein in the step d) possible combinations of local minima from the set the gray value gradients of all search directions by method step c) the combination that determines the normalized energy function is determined from the determined amounts of the gradients, the axial lengths and the ratio of the axial lengths of the resulting Contour minimized.
  8. Method according to Claim 1, in which the center point of the resulting contour is determined in method step e) and determined as starting point M n + 1 for the renewed method step b), in that the combination of local minima with the lowest total energies determined in method step d) determines the vertices form the contour.
  9. Process according to claims 1 to 8, at the resulting contour represents an ellipse.
DE102008018701A 2008-04-09 2008-04-09 Method for automated detection and segmentation of the papilla in fundus images Withdrawn DE102008018701A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
DE102008018701A DE102008018701A1 (en) 2008-04-09 2008-04-09 Method for automated detection and segmentation of the papilla in fundus images

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
DE102008018701A DE102008018701A1 (en) 2008-04-09 2008-04-09 Method for automated detection and segmentation of the papilla in fundus images
PCT/EP2009/002369 WO2009124679A1 (en) 2008-04-09 2009-04-01 Method for the automatised detection and segmentation of papilla in fundus images

Publications (1)

Publication Number Publication Date
DE102008018701A1 true DE102008018701A1 (en) 2009-10-15

Family

ID=40759002

Family Applications (1)

Application Number Title Priority Date Filing Date
DE102008018701A Withdrawn DE102008018701A1 (en) 2008-04-09 2008-04-09 Method for automated detection and segmentation of the papilla in fundus images

Country Status (2)

Country Link
DE (1) DE102008018701A1 (en)
WO (1) WO2009124679A1 (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE69731167T2 (en) 1996-12-03 2005-11-24 Nidek Co., Ltd., Gamagori Method and device for evaluating stereo images of the ocular fundus

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3594468B2 (en) * 1997-11-21 2004-12-02 株式会社ニデック Fundus image analysis method
US6031935A (en) * 1998-02-12 2000-02-29 Kimmel; Zebadiah M. Method and apparatus for segmenting images using constant-time deformable contours
JP4990143B2 (en) * 2004-09-21 2012-08-01 イメドース ゲーエムベーハー Method and apparatus for analyzing retinal blood vessels in digital images
US20070116338A1 (en) * 2005-11-23 2007-05-24 General Electric Company Methods and systems for automatic segmentation of biological structure

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE69731167T2 (en) 1996-12-03 2005-11-24 Nidek Co., Ltd., Gamagori Method and device for evaluating stereo images of the ocular fundus

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Tang, Yandong, et al; "Automatic segmentation of the papilla in a fundus image based an the C-V-model and a shape restraint"; Proceeding of the 18th International Conference an Pattern Recognition (ICPR'06); 0-7695-2521-0/06; 2006 IEEE
Xu, Juan, et al; "Optic disk feature extraction via modified deformable model technique for glaucoma analysis"; Pattern Recognition 40 (2007) 2063-2076

Also Published As

Publication number Publication date
WO2009124679A1 (en) 2009-10-15

Similar Documents

Publication Publication Date Title
Morales et al. Automatic detection of optic disc based on PCA and mathematical morphology
RU2706372C1 (en) System and method for measuring refraction error of eye based on subjective measurement of distance
US9149179B2 (en) System and method for identifying eye conditions
Lu Accurate and efficient optic disc detection and segmentation by a circular transformation
Aquino et al. Detecting the optic disc boundary in digital fundus images using morphological, edge detection, and feature extraction techniques
US9402538B2 (en) Photorefraction ocular screening device and methods
Annunziata et al. Leveraging multiscale hessian-based enhancement with a novel exudate inpainting technique for retinal vessel segmentation
Wong et al. Level-set based automatic cup-to-disc ratio determination using retinal fundus images in ARGALI
Kauppi et al. The diaretdb1 diabetic retinopathy database and evaluation protocol.
Deepak et al. Automatic assessment of macular edema from color retinal images
Yin et al. Automated segmentation of optic disc and optic cup in fundus images for glaucoma diagnosis
Almazroa et al. Optic disc and optic cup segmentation methodologies for glaucoma image detection: a survey
Lowell et al. Optic nerve head segmentation
Niemeijer et al. Segmentation of the optic disc, macula and vascular arch in fundus photographs
US6095989A (en) Optical recognition methods for locating eyes
US5868134A (en) Retinal disease analyzer
JP5955163B2 (en) Image processing apparatus and image processing method
US6714672B1 (en) Automated stereo fundus evaluation
Mary et al. An empirical study on optic disc segmentation using an active contour model
Joshi et al. Optic disk and cup segmentation from monocular color retinal images for glaucoma assessment
Lim et al. Integrated optic disc and cup segmentation with deep learning
Yin et al. Vessel extraction from non-fluorescein fundus images using orientation-aware detector
Matsopoulos et al. Multimodal registration of retinal images using self organizing maps
US7614745B2 (en) System for analyzing eye responses to automatically determine impairment of a subject
Sekhar et al. Automated localisation of retinal optic disk using Hough transform

Legal Events

Date Code Title Description
OR8 Request for search as to paragraph 43 lit. 1 sentence 1 patent law
R005 Application deemed withdrawn due to failure to request examination