Upload
sreeja-ganesh
View
37
Download
6
Tags:
Embed Size (px)
DESCRIPTION
Optic disc detection and extraction of features.
Citation preview
Image Processing
Medical Image Processing310912405023
G.SreejaProject Guide: Mr.S.Sadagopan
(Assistant Professor)(Computer Science and Engineering Department –
Jerusalem college of engineering)
OPTIC VESSEL AND DISK DETECTION IN RETINAL IMAGES
OBJECTIVE
The vessel detection and disc extraction from the retinal image using effective algorithm for analyzing the diabetic retinopathy severity.
Fast, Reliable and Efficient Optic Disc (OD) Localization
and segmentation are important tasks in automatic eye disease screening.
Optic Disc location arrived using Template Matching and the template is adaptable for different image resolutions.
By Initializing the Optic disc center and OD Radius, a Fast Hybrid Level set model combines the OD region and local gradient information to the segmentation of the disc boundary.
Morphological Filtering brings the removal of blood vessels and bright region other than Optic disc.
ABSTRACT
INTRODUCTION AND MOTIVATION
Analysis of the retinal image using reliable and efficient algorithm is necessary .
The analysis includes optic vessel and disc extraction effectively. Features taken from the retina gives the information about retinal abnormalities.
80% of the abnormalities includes for the eye is caused by diabetic retinopathy and glaucoma. So it is a major concern.
LITERATURE SURVEYTitle Method used Advantage Disadvantage Parameters used
H. Yu, Member et.al Fast localization and segmentation of optic disk in retinal images using directional matched filtering and level set IEEE transactions on information technology in bio medicine Volume 16 No.4 2012
Localization and Segmentation using direct matched filtering and level sets
Robust and efficient No accuracy in advanced stage.Large milineated severe PPA.Constant Threshold value
Image Foot PrintRadius of optic diskOverlapping RatioMADHausdroff
Sandra Morales et.al Automatic Detection of Optic Disc Based on PCA &Mathematical MorphologyIEEE transactions on Medical imaging volume 32 No.4 2013.
PCA and Mathematical Morphology
Full automation of algorithmDoesn’t required clinical interference
Perfect Segmentation is difficult
Jaccards CoefficientDies AccuracyTPF and FPFMAD
Xiayu Xu et al Vessel Boundary Delineation on Fundus ImagesUsing Graph-Based Approach, IEEE transactions on Medical imaging volume 30, No.6 2011
Boundary delineationGraph based on approach
Few database closed to reference standardCould find retinal vessel boundary of even small vessels boundary
Constant value for thresholdSigma Value for constant hence not reliable
MeanStandard deviationGaussian Derivatives
LITERATURE SURVEY (Continuation)
Title Method used Advantage Disadvantage Parameters used
Marvin Tell Alonso et.al Edge Enhancement Algorithm Based on the Wavelet Transform for Automatic Edge Detection in SAR Images IEEE Transactions on GEO science and remote sensing volume 49, No.1 2011
Edge EnhancementWavelet transformationAutomatic edge detection
Robust and effective for applicationsNo Pre Filtering
UnsupervisedEdge Enhancement Critical
Hausdroff distancePfp PNF
Keith A. Go atman et al Detection of New Vessels on the OpticDisc Using Retinal PhotographsIEEE transactions on medical imaging volume 30 No 40 2011
Detection vessels New growing vessels were detected
Requires every possible referral features to be detectedReliably system to be safe
Ansari Bradley TestWater Shed TransformKthresh
Accurate extraction of vessel features so that we get
the finer details of vessels. Exact Optic disc location in case of severe cases.
PROBLEM STATEMENT
Usage of constant threshold value. Detection of vessels not accurate The methods used are not fast and reliable Perfect segmentation is difficult
DRAWBACKS OF CURRENT APPROACH
ISSUES GOING TO BE ADDRESSED
Exact location of optic disc Many features of vessels to be identified
ARCHITECTURE DIAGRAM
OPTIC DISC SIZE ESTIMATION
BACKGROUNDNORMALISATION
TEMPLATE MATCHING
DIRECTIONAL MATCHED FILTERING
MODEL PARAMETER OPTIMIZATION
SATURATION DETECTION IN RED
CHANNEL ROI
BLOOD VESSEL REMOVAL
BRIGHT REGION REMOVAL
FAST, HYBRID LEVEL SET SEGMENTATION
OPTIC DISC LOCALIZATION
LEAST SQUARE ELLIPSE FITTING
INPUT IMAGE
OD LOCATION
OPTIC DISC SEGMENTATION
OUTPUT IMAGE
EXTRACTION OF OPTIC VESSEL
AND ITS PARAMTER
OPTIMIZATION
AUTOMATIC OPTIC DISC
SIZE ITERATIVE METHOD
MODULES SPLIT UP
OPTIC DISC DETECTIONPreprocessing Template MatchingSegmentation Exact detection of Optic DiscDisc features extraction
OPTIC VESSEL DETECTIONPreprocessing Edge detection methodExact detection of Optic VesselVessel feature extraction
Background Normalization: To reduce the false detection of
OD candidates due to non-uniform illumination, we applied
an image illumination correction using Histogram
equalization method and we get an normalized image with
over smoothed background.
Template Matching: Is a binarization technique where the
image takes value 1 and background 0.Pearson coefficient
is used to get the values where values above 0.5 are taken
and rest are ignored. The value ranges from -1 o 1.The
formulae used to calculate is as follows
DATA FLOW DIAGRAM
BACK GROUND NORMAILISATION
TEMPLATE MATCHING
MATCHED FILERINGPEARSON COEIFFICENT
ESTIMATION
BINARIZATION
MATCHED FILERING
EXUDATES REMOVAL
REMOVAL OF FALSE POSITIVE
1DFD 2DFD
OD Segmentation:
In this stage, we first detect the saturation in the red channel of
the image.
In the red channel, the OD often appears with the most contrast
against the background, while vessels appear less prominently.
Thus, the OD segmentation algorithm is performed in red channel.
Then we remove the blood vessel from the retinal image because
Interference of blood vessels is one of the main difficulties in
accurate OD boundary segmentation.
In the OD segmentation ROI, areas containing features with bright
pigmentation, such as choroid vessels, exudates, and cotton wool
spots may interfere with the OD boundary segmentation.
We use morphological reconstruction to suppress the bright
regions that are lighter than their surroundings and are also
connected to the image border.
The new hybrid level set model uses speed function ,edge and
step function to get partial differential equation . The
threshold value is used to estimate region of interest which
depends on mean and standard deviation values.
Speed function is denoted as
The step and the edge function is used to get the narrow band region. The following equations are used.
The threshold value λ is used to get the region of interest. Where μ Mean c coefficient and σ is the standard deviation.
Exact Detection of optic discThe Disc curvature is used to define an internal force to make the evolving contour smooth during the hybrid level set model deformation. The final curve may still appear irregular due to the influence of strong blood vessels. To provide for a smooth contour, we fit the segmented OD boundary with an ellipse using the least-squares optimization. This step generates smooth OD borders that can be used for cup-to-disk ratio computation.
Vessel Feature ExtractionThe Features of the vessel such as segment length ,segment gradient ,direction of vessel growth are extracted. Features of normal and abnormal eyes are extracted.
Let F be the speed funcion which depends on local and global properties.Local properties = Normal direction and curvatureGlobal properties = shape, position and other independent external facts.To get the narrow band edge and step function are used.
We arrive at the curve evolution partial differential equation as follows
ALGORITHM – OPTIC VESSEL AND DISC DETECTION
∂ϕ∂t = gεk|∇ϕ| + β1(1 − λ)|∇ϕ| + β2∇g · ∇ϕ.
Based on additive operative approach the PDE equation
simplified as∂ϕ∂t = α(1 − λ) + β div(g∇ϕ)Λ – Threshold based on mean standard deviation and coefficientInitial threshold is set. Below he threshold values are ‘x’ and
above the threshold values are y.CalculationX / (total no of threshold) + Y / (total no of threshold) = zNew threshold = z/2Again The new threshold is calculated and region of interest is
arrived.
1] D. Pascolini and S. P.Mariotti, “Global estimates of visual impairment: 2010,” Br. J. Ophthalmol., pp. 614–621, 2011. [2] World Health Org., Action plan for the prevention of blindness and visual impairment 2009–2013 2012. [3] H. R. Taylor, “Eye care for the community,” Clin. Exp. Ophthalmol., vol. 30, no. 3, pp. 151–154, 2012. [4] T. Walter, J. C. Klein, P. Massin, and A. Erginay, “A contribution of image processing to the diagnosis of diabetic retinopathy-detection of exudates in color fundus images of the human retina,” IEEE Trans. Med. Imag., vol. 21, no. 10, pp. 1236–1243, Oct. 2012. [5] L. D. Hubbard, R. J. Brothers, W. N. King, L. X. Clegg, R. Klein, L. S. Cooper, A. Sharrett, M. D. Davis, and J. Cai, “Methods for evaluation of retinal microvascular abnormalities associated with hypertension/ sclerosis in the atherosclerosis risk in communities
References