AUTOMATIC DETECTION OF GLAUCOMA

CONTENTS
1. CAUSES OF GLAUCOMA
2. IDENTIFICATION OF GLAUCOMA
A. calculation of CDR
B. ISNT rule for identification of neuro retinal rim
C.papillary atrophy detection
D. Hemorrhage detection.

A. calculation of CDR

Steps involving for calculation of cdr as follows:

Green channel extraction

Removing of noise present in the image using median filtering
Contrast enhancement for better visibility of the image using histogram
equalization
Removing the blood vessels using gray level morphological operation.
Automatic localization of optic disc.
Identification of optic disc center using circular hough transform after edge
detection
Segmentation of optic disc will be found using applying deformable active
contours.
From the segmented optic disc binary image we need to calculate features of
optic disc like area
Calculation of optic cup is utilizing the previously segmented optic disc
Initially need to identify the vessel bends within the optic disc area and also
consider the pallor region of the disc by using these two steps we can able to
identify the cup area(segmentation of cup)

B. identification of neuro retinal rim (ISNT)

The Inferior, Superior, Nasal and Temporal quadrants for retinal image Inferior quadrant

is the lowermost quadrant and superior quadrant is the uppermost

Quadrant. The nasal quadrant is the right quadrant in right eye and left quadrant in left

eye. The temporal quadrant is the left quadrant in right eye and right quadrant in left eye.
The relevance of these quadrants in fundus image analysis is that the segmented fundus

eye features in the ISNT quadrant scan be used for the calculation of ISNT ratio.
The ISNT rule states that n normal eyes the thickness of the neuroretinal rim along the
Cardinal meridians of the optic disc ,that is the rim width, decreases in the order inferior
(I), superior (S), nasal (N),temporal (T), and that the neuroretinal rim in glaucomatous
optic disc violates this quantitative relationship. The four variants of ISNT rule are (I>S,
I>S>T, I>T, and combined I>T and S>T).

To calculate ISNT ratio we need to generate the mask as follows

The mask image for each quadrant is generated exploiting the characteristics of
array and the centroid value. The mask image act as an image filter thus the mask
image should be such that the desired quadrant should be white and all other
quadrants should be black.

Inferior quadrant is the lowermost quadrant. The mask for this quadrant should be
such that inferior quadrant is white or binary 1 and all other quadrants are black or
binary 0.

Superior quadrant is the uppermost quadrant and the mask image should be such
that superior quadrant is white or binary 1 and all other quadrants are black or
binary 0.

The leftmost quadrant is the nasal quadrant for left eye and temporal quadrant for
the bright eye. The mask for left quadrant has white pixels on left quadrant and
black pixels on other quadrants.

The rightmost quadrant is the temporal quadrant for left eye and nasal quadrant
for the right eye. The mask for right quadrant has white pixels on right quadrant
and black pixels on other quadrants.

C. papillary atrophy detection using GLCM features.

Texture analysis of an image is the study of mutual relationship among intensity

values of neighboring pixels repeated over an area larger than the size of the

relationship. The texture features considered for this work are mean, standard
deviation (SD), third moment, entropy and homogeneity. Co-occurrence matrix
defines the spatial distribution of gray level from which homogeneity feature can
be acquired.

D. Hemorrhage detection.

Step1: Image enhancement.

We initially work on the color image; contrast of original color image is increased so that
hemorrhages are highlighted. Then median filter, which is a nonlinear filter, is used to
reduce salt and pepper noise.
Step2: Image Subtraction
Color image enhanced in previous step is converted to gray image. Average width of the
blood vessels is reported in the literature as 250m. Median filter of size more than twice
as the blood vessel width is used and gray image is subtracted from median filtered
image.
Step3: Image thresholding and strengtning
By selecting suitable threshold image is converted to binary form which shows only
blood vessels and hemorrhages. Image strengthening is done by setting pixel to 1 if five
or more pixels in its 3-by-3 neighborhood are 1's; otherwise, set the pixel to 0.
Step4: Image thinning, Erosion and skeletonization
It removes pixels so that an object without holes shrinks to a minimally connected stroke,
and an object with holes shrinks to a connected ring halfway between each hole and the
outer boundary. Further erosion is performed with disc structuring element of size 6.With
this operations blood vessels are thinned. To further suppress blood vessels we use
skeletonization.

Step5: Morphological closing, Image complement and Component Labeling

The morphological closing operation is performed to suppress blood vessels completely

and then to get actual size of the hemorrhages. Image obtained after closing operation is
complemented and component labeling is done so as to count number of hemorrhages
and also tofind area occupied by hemorrhages.