Docstoc

Early Detection of Glaucoma in Retinal Images Using Cup to Disc Ratio fundus

Document Sample
Early Detection of Glaucoma in Retinal Images Using Cup to Disc Ratio fundus Powered By Docstoc
					    Early Detection of Glaucoma in Retinal Images
               Using Cup to Disc Ratio
                             S.Kavitha1, S.karthikeyan2, Dr.K.Duraiswamy3
1
  Asst.Prof.Nandha Engineering College, Erode, 2 Professor, K.S.R.College of Engineering, Tiruchengode.
3
  Dean, K.S.R.College of Technology, Tiruchengode. email:gsk_kavitha@rediffmail.com
Abstract- Glaucoma is a disease characterized by elevated           Optical Coherence Tomography (OCT) and Heidelberg
intraocular pressure (IOP). This increased IOP leads to             Retinal Tomography (HRT) are very expensive [1, 2]
damage of optic nerve axons at the back of the eye, with            compared to digital fundus images. With the help of image
eventual deterioration of vision. CDR is a key indicator for        processing, the features of the fundus images such as optic
the detection of glaucoma. The existing approaches                  disc and cup could be localized to suspect the glaucoma [4, 5].
determined the CDR using manual threshold analysis                  The detection of OD position is a pre-requisite for the
which is fairly time consuming. This paper proposes two             computation of some important diagnostic indexes like
methods to extract the disc automatically. The component            glaucoma. The optic disk is the region on the retina at which
analysis method and region of interest (ROI) based                  optic nerve axons enter and leave the eye. The optic disc and
segmentation is used for the detection of disc. For the cup,        cup were located by identifying the area with the highest
component analysis method is used. Later the active                 variation in intensity of adjacent pixels [6]. This paper
contour is used to plot the boundary accurately. This               presents the automatic detection of optic disc by region of
method has been tested on numerous image data sets from             interest based segmentation and component labeling. As the
Madurai Eye Care Centre, Coimbatore.                                shape of the optic disc is important to diagnose eye diseases,
                                                                    the exact boundary detection of the optic disc is investigated.
                                                                    “Active Contour” was applied to detect the exact contour of
Index terms - Glaucoma, fundus, Region of Interest (ROI),           optic disc. The major advantage of this method is their ability
Component Analysis.                                                 to bridge the discontinuities in the image features being
                                                                    located. The contour was plotted by calculating the centroid of
                                                                    the image feature. Later the optic cup has been detected by
                       I .INTRODUCTION                              component analysis method. Then the number of white pixels
                                                                    was calculated to find the area of disc and cup from the result
   Glaucoma is defined as „multi factorial optic neuropathy‟        obtained by the above mentioned methods. The estimation of
which is a potentially blinding disease which affects 66.8          boundary was achieved by active contour. With the help of
million people worldwide. It is the second leading cause of         detected area of disc and cup the CDR was calculated to
blindness. Risk assessment of the disease goes a long way in        suspect the glaucoma. Finally clinical CDR is compared with
diagnosis and management of the disease. Your eye has               the proposed three methods.
pressure just like your blood and when this intra ocular
pressure (IOP) increases to dangerous level, it affects the optic                     II. METHODOLOGY
nerve. Although the raised IOP is a significant risk factor for
developing glaucoma, there is no set threshold for IOP that         A. Extraction of Optic Disc
causes glaucoma. This can result in decreased peripheral
vision and eventually leads to blindness. Glaucoma is similar       1. Manual Threshold Analysis: The formulation of the Manual
to ocular hypertension, but with accompanying optic nerve              threshold for extracting the optic disc includes the removal
damage. Vision loss is caused by damage to the optic nerve,            of the blood vessels in the retinal images. The
which carries image information from the light receptors to the        morphological operation such as the dilation, erosion, is
brain. Untreated glaucoma leads to permanent damage to the             performed. The morphological functions are applied to do
optic nerve and result in visual field loss. Early detection and       the pre-processing. The dilation and erosion of A and B is
prevention is the only way to avoid total loss of vision.              defined by,
Automated screening for glaucoma associated eye pathology
is of benefit for several reasons. The most important of these           A∙B = ( A    B) ΘB                                    (1)
is that pathology can be detected at the asymptomatic stage of
disease progression and in amendable to treatment with good         Dilation causes objects to grow in size by adding pixels to the
outcomes. The ailment is physiologically described as the           boundaries of the object in the image. Erosion is done to
degeneration of optic nerve cells with associated visual field      contrast the boundary of the object. The result of this
effects. The suspection of glaucoma makes easier by means of        operation has smooth image without any blood vessels.
computer screening [3]. The detection of glaucoma through           Individual pixels in a grayscale image are marked as „object‟
pixels if their value is greater than some threshold value            B. Extraction of Optic Cup
(assuming an object to be brighter than the background) and as        Component Analysis Method: The segmentation of optic cup
„background‟ pixels otherwise. Typically, an object pixel is          is considerably more challenging than the optic disc due to
given a value of „1‟ while a background pixel is given a value        high density of vascular architecture traversing the cup
of „0‟. The minimum threshold value of the disc is set in order       boundary. The disc and cup could not be easily distinguished
to extract the optic disc boundary. The boundary is estimated         as the border between the two was unclear. The component
by plotting the boundary over the input image as in figure().In       analysis method will localize the optic cup efficiently, even
most conventional level set method, undesired effects also            though the image is low contrast. In this paper, the green
known as shocks can develop during the evolution of the level         component from the original fundus image is extracted in
set function, resulting in inaccurately detected contours. To         order to detect the cup [11]. The morphological operation like
avoid this, periodic re-initialization of the level set function is   closing and opening are performed to get the area of cup more
necessary but may not be an optimum solution.                         accurately [6]. The close operation would fill the gap of the
                                                                      cup and also smoothen its boundary. The open operation
2. Color Component Analysis: In the component analysis                would remove any small stray bright spots that are present in
method, RGB components are analyzed and it was found that             the image [12]. The component analysis method detects the
the optic disc was more easily discriminated in RED image             area of the cup more accurately than the manual threshold
[11]. In order to measure disc more accurately, the blood             analysis. Later the result of the morphological image is
vessels in the image had to be removed. This is achieved by           converted to binary image. The exact optic cup was obtained
performing the closing and opening operation. Closing is              from the binary image. By calculating the number of white
similar to dilation and it tends to enlarge the boundaries of         pixels in the binary image obtained, the area of cup was
foreground regions in an image and shrink background color            measured.
holes in such regions. Opening is similar erosion and it tends
to remove some of the foreground pixels from the edge of
region of foreground pixels [6]. First CLOSE operation is                           III. DETERMINATION OF CDR
done followed by the OPEN [12]. The boundary of the optic
disc is detected by converting the opening image into the               There are many features for identifying the vision impaired
binary image in which white pixels closely approximate the            disease glaucoma. The first and important feature is the cup to
edge of the optic disc. This method accurately determines the         disc ratio, which specifies the change in the cup area. Due to
boundary of the optic disc only when the image has the                glaucoma the cup area will increase slowly by intra ocular
contrast disc. If the disc boundary is blurred then it fails to       pressure (IOP) and results in dramatic visual loss. Increase in
extract the boundary exactly.                                         cup area is analyzed by examining the CDR value. The CDR
                                                                      was calculated by taking the ratio between the area of optic
3.ROI based segmentation: The optic disc is extracted by              cup and disc. If the CDR > 0.3 indicates the suspection of
finding the region of interest based on colors. In order to           glaucoma. If the CDR< 0.3 , it is considered as normal image.
calculate the ROI, in the original image the mathematical
morphology like dilation, erosion is done to smoothen the
image. After performing the morphological functions the
small holes gets filled and object boundary is smoothen. The
ROI for the input image is then examined; the result of this
step gives the binary image of the optic disc. To localize the
boundary exactly the component labeling is used. In this
image areas are label by using the neighborhood connecting
pixels. All the connected pixels with the same input value are
assigned the same identification label. Figure (1) shows the
component labeling algorithm.



                                                                                      (a)                           (b)
                                                                                      Fig 2.suspection of glaucoma
                                                                                       (a) CDR<0.3 (b) CDR>0.3



                     (a)                   (b)
        Fig.1. Result from the component labeling
        (a) Before labeling (b) After labeling.
            IV. EXPERIMENTAL RESULTS
                                                                 component analysis for the cup. In third and last row the
   The first row presents the input image and the next row       cup was same, because in both component analysis method
shows the results of manual threshold analysis. The third        was used. Table 1 compares the CDR value obtained from
row presents simulation results of component analysis for        manual threshold analysis, component analysis and finally
cup and disc. The last row gives the result obtained from        with ROI based segmentation for disc and component
ROI based segmentation method for disc and                       analysis method for cup.




                          Fig 3.First row: Input images ; Second row: Result from manual threshold;
                         Third row: Result from component analysis; Fourth row: Result from ROI method.
The table relates the clinical CDR with proposed CDR value. It
can be obviously seen that the ROI and component analysis                                                                                              ROI and Component Analysis
method provides better result when compared with the other
two methods. Then graph is plotted for the clinical CDR with                                                                               0.8

the detected CDR for the three methods.                                                                                                    0.7




                                                                                                                       Cup to Disc Ratio
                                                                                                                                           0.6
Table 1: Comparison of clinical CDR value with several methods
                                                                                                                                           0.5
                                                                                                                                                                                       cdr.roi
                                                    MANUAL               COMPONENT            ROI &                                        0.4
                                CLINICAL            THRESHOLD            ANALYSIS             COMPONENT
IMAGES                          CDR                 CDR                  CDR                  CDR                                          0.3
                                                                                                                                                                                       cdr.clinical

                                                                                                                                           0.2
G1                                       0.35                  0.4616                0.4733               0.3174
G2                                       0.35                 0.41668                0.3267               0.3251                           0.1

G3                                       0.25                  0.2689                0.2584               0.2216                            0
G4                                        0.5                  0.4013                0.7523               0.4956                                 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
G5                                       0.35                  0.6794                0.4236               0.3331
G6                                       0.35                  0.6041                0.5011               0.3712                                              Images
G7                                       0.35                  0.5136                0.4389               0.3543
G8                                        0.4                  0.5887                0.4889               0.3627
G9                                        0.3                   0.601                0.3006               0.3189
G10                                      0.55                  0.5892                0.5399                0.506      Fig.6. Comparison of clinical CDR and CDR from
G11                                       0.7                  0.5679                0.6912               0.6104            ROI method
G12                                      0.45                  0.5817                0.5297               0.4877
G13                                      0.35                   0.512                0.4188               0.3272
G14                                      0.35                  0.4689                0.4298               0.3638                         IV. POSTULATION
G15                                       0.2                   0.245                 0.244               0.2073
                                                                                                                       This work proposes the recognition of Glaucoma in
                                                                                                                   advance. There are several methods of treatment available to
                                                                                                                   impede progression of this disease. It is important to diagnose
                                                        Manual T hreshold
                                                                                                                   Glaucoma as early as possible to minimize the damage to optic
                          0.8
                                                                                                                   nerve head. The detection of Glaucoma by Optical Coherence
                          0.7
                                                                                                                   Tomography (OCT) and Heidelberg Retinal Tomography
      Cup to Disc Ratio




                          0.6                                                                                      (HRT) was used to differentiate between the glaucoma and
                          0.5
                                                                                              cdr.thres
                                                                                                                   non-glaucoma eye using neural network is very expensive. But
                          0.4                                                                                      in this approach, the funds photograph which is less expensive
                          0.3                                                                 cdr.clinical         is used. The Promising result suggests good potential for use of
                          0.2
                                                                                                                   ROI and component analysis in detection of Glaucoma in early
                          0.1

                           0
                                                                                                                   stage. The accuracy of classification also depends on the
                                 1 2     3 4    5 6     7 8 9 10 11 12 13 14 15                                    environment lighting conditions. Hence this method provides
                                                      Images                                                       an additional tool for physician to verify their diagnosis.

                                                                                                                                          V.CONCLUSION
      Fig.4. Comparison of clinical CDR and CDR from                                                                        The algorithms for the Advance identification of
             manual method                                                                                         Glaucoma by estimating CDR were developed in this paper.
                                                                                                                   ROI based segmentation is proposed to localize optic disk,
                                                    Component Analysis                                             which is estimated by using contour method, exactly when
                                                                                                                   compared with other methods even though the image is in low
                          0.8                                                                                      contrast. The optic cup was segmented using the component
                          0.7                                                                                      analysis and the threshold methods separately. The
      Cup to Disc Ratio




                          0.6                                                                                      performance of various methods was evaluated using the
                          0.5
                                                                                               cdr.com             proximity of the calculated CDR to the clinical CDR. It was
                          0.4

                          0.3                                                                  cdr.clinical
                                                                                                                   found that ROI, combined with the component analysis method
                          0.2
                                                                                                                   provides the better estimation of CDR. The implementation of
                          0.1                                                                                      the above said method would be more fruitful with the
                           0                                                                                       availability of more suitable data. The Algorithm proposed in
                                 1   2   3 4    5   6    7   8 9 10 11 12 13 14 15
                                                                                                                   this paper has high value in clinic practice for automatic
                                                        Images                                                     screening of early diagnosis of Glaucoma. This proposed
                                                                                                                   method can be used as an adjunct tool by the physicians to
        Fig.5. Comparison of clinical CDR and CDR from                                                             cross check their diagnosis.
               component analysis method
                   ACKNOWLEDGMENT                                  [8] N.Inoue, K.Yanishima, K. Magatani, and T.A.K.T
                                                                        Kurihara, “Development of a simple diagnostic method for
           The authors would like to thank Dr.Srinivasan from           the glaucoma using ocular Fundus pictures,” in
Madurai Eye Care Hospital, Coimbatore for his suggestions               proceedings of the 2005 IEEE Engineering in Medicine
and for supplying the fundus Photographs for our tests.                 and Biology 27th annual international conference , pp,
                                                                        3355 -3358, sep 2005.
                                                                   [9] H. Li and O.Chutatape, “A model – based approach for
                        REFERENCES                                      automated feature extraction in fundus images,” in
                                                                        proceedings of the Ninth IEEE International conference
[1] Acharya, U. R., Ng, E.Y.K., and Suri, J. S., “Image                 on computer vision ,vol .1, pp 394-399,oct 2003.
     modeling of human Eye” Artech House, MA, USA, 2008a,          [10] M. Kass, A. Witkin and D. Terzopoulos, “Snakes: Active
     April.                                                             contour models,” Int. J. Computer vision, vol. 1, pp 321-
[2] U.R., Chau, K. C., Ng, E. Y. K., Wei, W., and Chee, C.,             331, 1988.
     “Application of higher order spectra for the identification   [11] Jagadish Nayak. Rajendra Acharya U. P.Subbanna Bhat.
     of diabetes retinopathy stages”. J. Med syst. USA. 2008b.          Nakkul Shetty. Teik –Cheng Lim, “Automated Diagnosis
     doi: 10.1007/s10916-008-9154-8.                                    of Glaucoma Using digital Fundus Images,” June 2008.
[3] Song, X., Chen, Y., Song, K., and Chen, Y., “A Computer        [12] Walter, T., Klein, J. C., Massin, P., and Erginay,A.,“A
     – based diagnosis of glaucoma using an artificial neural           Contribution of image processing to diagnosis of diabetic
     network”. Proceedings of 17th Annual Conference IEEE               retinopathy – detection of exudates in color fundus images
     Engineering in Medicine and Biology, 1, 847-848, 1995.             of the human retina”. IEEE trans. Med. Imaging.21:1236-
[4] Viranee Thongnuch, Bunyarit Uyyanonvara, “Automatic                 1243,2002. doi: 10.1109/TMI.2002.806290.
     optic disc detection from low contrast retinal images of      [13] J. Liu, D. W .K Wong, J. H. Lim, X. Jia, F. Yin, H. Li, W.
     ROP infant using mathematical morphology”, 2000.                   Xiong, T.Y. Wong, “Optic Cup and Disc extraction from
 [5] Nayak, J. Bhat, P.S., Acharya, U. R., Lim, C.M., and               Retinal Fundus Images for Determination of Cup- to- Disc
     Kagathi, M., “Automated identification of different stages         Ratio”, in proceedings of 2008 IEEE Engineering pp 978-
     of diabetic retinopathy using digital fundus images”. J.           1-4244-1718-6/08/ .
     Med. Sys.USA. 32:2107-115, 2008, doi:10.1007/s10917-          [14] Giri Babu Kande, T.Sathya Savthri, P.Venkata Subbaiah,
     007-9113-9.                                                        M.R.N Tagore, “Automatic Detection and Boundary
 [6] Wong, L.Y., Acharya, U.R., Venkatesh, Y.V., Chee,                  Estimation of Optic Disc in Fundus Images using
     C.Lim, C.M., and Ng, E. Y. K., “Identification of different        Geometric Active Contours,” Feb 2009.
     stages of diabetic retinopathy using retinal optical
     images.” Inf.sci. May 2008.
[7] J. A. Sethian, “Level Set Methods”, Cambridge
     Monographs        on    Applied     and     Computational
     Mathematics”, Cambridge, 1996.

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:108
posted:9/9/2010
language:English
pages:5
Description: Early Detection of Glaucoma in Retinal Images Using Cup to Disc Ratio fundus