Ecs10_contrib

8/3/2019 Ecs10_contrib

1/6

AUTOMATIC MACULA DETECTION IN HUMAN EYE FUNDUS AUTO-

A DNAN R ASHID C HAUDHRY 1 , C AREN B ELLMANN 2 , VALERIE L E T IE N 3 , JEA N -C LAUDEK LEIN 1 AN D E STELLE PARRA -D ENIS 11Mines Paris Tech, Centre de Morphologie Math ematique, 35 rue Saint Honor e 77305 Fontainebleau, France,2Institut Curie, 26 rue Ulm 75005 Paris, France, 3 Centre Hospitalier Intercommunal de Cr eteil, 40 avenue deVerdun 94010 Cr eteil, Francee-mail: [email protected], [email protected], [email protected], [email protected], [email protected]

ABSTRACT

Fundus AutoFluorescence (FAF) images are widely used in the diagnosis and follow-up of Age-relatedMacular Degeneration, which is the leading cause of blindness in people over 55. There are two kinds of AMD: wet and dry. The most common is the dry form. It is characterized by atrophies of the the retinalpigment epithelium (RPE) with subsequent photoreceptor degeneration. The atrophy severity depends on itssize and its location with regard to fovea. The fovea is the macula center, which is the retina central zone(about 2mm of diameter). The macula has a high density of cone photoreceptors and is not supplied by theretinal capillaries; it receives its blood supply from the choroid. For the determination of the lesions location,we developed an automatic positioning and orientation of a retinal grading grid. Centered on the fovea, itprovides information on the spatial distribution of pathologies and thus helps to determine the severity and thedanger associated to the detected retinal lesions. This automatic positioning is based on a top down approach,which starts with the vascular tree segmentation and skeletonization. The detected vessels help in localizationand detection of the Optic Disc (OD). An approximate OD centre is found by ellipse tting to the boundary of OD. The vascular tree and the ellipse tting help in detection of the macula and its center the fovea in relationto the OD center.

Keywords: Retinal images, Grading grid, Fundus autouorescence image.

INTRODUCTION

Fundus AutoFluorescence (FAF) imaging is a fastand non-invasive technique developed over the last tenyears for studying the retina. It uses the uorescentproperties of a pigment, called Lipofuscin (LF),located in the Retinal Pigment Epithelum (RPE) cells.It allows us to study the health and viability of the RPE(Schmitz-Valckenberg et al. , 2008; Bellmann et al. ,2003). The images obtained with FAF techniquescorrespond to LF distribution in the RPE. The normalFAF pattern (Fig. 1a) shows a higher level of LF inthe parafoveal area and a decrease of LF toward theperiphery. It decreases in the macular zone with alowest value corresponding to the fovea (ie the maculacenter which is the retina central zone measuring about2mm of diameter). This decrease corresponds to theabsorption of the short wavelength light by melaninand luteal pigment and also partly to decreasedLF content in that area. There is also a reducedautouorescence at the Optic Disc (OD) due to theabsence of the RPE and at the blood vessels due to themasking of the RPE by the blood circulation.

Retinal pathologies in FAF images correspondto a FAF abnormal increase resulting from LF

accumulation, or a FAF decrease or a FAF lack (forexample RPE atrophies Fig. 1b). Thus FAF imagingmay provide ideas about the RPE metabolic state.It represents a useful diagnostic and prognostic toolfor ophthalmologists to identify specic FAF changeshaving been associated with retinal disease.

In this paper, the studied FAF images areacquired with an angiograph equipped with a ScanningLaser of 488 nm wavelength and an appropriate barrierlter (Scanning Laser Ophthalmoscope). The images,centered onto the fovea, correspond to a 30eld of view, which encompasses almost the posterior pole of the retina (Fig.1). Their size is 768 768 pixels.

This paper is organized as follows. The rstpart is dedicated to the automatic detection of themacular zone by a top-down approach. This approachstarts with the vascular tree segmentation and theskeletonization, the detected vessels help with the ODlocalization and detection. The vascular tree and theOD help with the detection of the macular zone. Thesecond part deals with the automatic grading gridsetting in two steps. The rst step is the detection of

the macular zone center to place the grid center. Thesecond step is the estimation of the OD center by usingan ellipse tting algorithm to orientate the grading

1

S t e r e o

l o g y a n d

I m a g e

A n a

l y s i s .

E c s

1 0 -

P r o c e e d

i n g s o

f t h e

1 0 t h E u r o p e a n

C o n g r e s s o

f I S S

,

( V . C

a p a s s o e t a

l . E d s .

) , T h e

M I R I A M

P r o

j e c t

S e r

i e s ,

E S C U L A P I O

P u

b .

C o . ,

B o

l o g n a ,

I t a

l y ,

2 0 0 9

FLUORESCENCE IMAGES: APPLICATION TO EYE DISEASELOCALIZATION


2/6

C HAUDHRY A DNAN et al. : Automatic macula detection

grid. The third part concludes the paper and looks atthe prospective works.

(a) (b)

Fig. 1. Example of typical FAF image obtained for anormal eye (a), and for a pathologic one (b).

AUTOMATIC DETECTION OF THEMACULAR ZONE

Fig. 2. The top-down approach used to determine themacular zone.

This section presents the automatic detection of the macular zone. First a state of the art is provided.Second we present an original method to automaticallydetect the macular zone by a top down approach (Fig.2).

MACULA DETECTION: STATE OF THEART

In the literature there are few papers dealingwith the macular localization in FAF modality. In(Sinthanayothin et al. , 1999), on color images, atypical template of the macular region is formedand correlated with the image to localize thefovea in the OD neighborhood. On the uoresceinangiography, mathematical morphology and regiongrowing algorithms are applied for the macular zonelocalization in (Zana et al. , 1998); in (Gutierrez

et al. , 2000), an active contour is performed basedon B-snakes with greedy minimizing algorithm tocharacterize the foveal zone boundary. In (Li and Opas,

2004) an automated extraction of the macula on colorimages is performed by detecting the main vessels bya modied active shape model. The latter is tted witha parabola to specify the candidate region. The fovea isthen located by a clustering of the darkest pixels as themacular candidate region. In (Tobin et al. , 2007), theparabolic model of a vascular tree is used to estimatethe fovea position. Our approach is derived from thesetwo last presented methods.

MACULA DETECTION APPROACH

We propose an original method developed in twostages: the macular zone detection and localization.This method is based on the hypouorescence of the macular region in FAF images. The use of itsposition with regard to the vessel endpoints and the ODdenes a mask region which contains the macula. Thistwo-step method gives a robust technique in normalsubjects as well as in patients with retinal disease.

Extraction of the vessel endpoints In FAFimages, the blood vessels are the most distinctivefeature, they appear as a tree like structure originatingfrom the OD. Their characteristics are: a diameterof less than 15 pixels, an elongated piecewise linearshape and a gray level darker than the background.In the literature, the retinal blood vessel segmentationalgorithms are categorized into a wide variety of techniques reviewed in (Kirbas and Quek, 2004).We have designed a fast and robust algorithmto extract the vessels on FAF imaging based onmathematic morphology (Serra, 1982). It is dividedinto 5 steps (illustrated on Fig. 3): preprocessing,vessel segmentation, skeletonization, post ltering andrened vessel skeleton.The preprocessing step consists of ltering the FAFimage with a gaussian lter (of size 5 and standarddeviation 1 .2) followed by a morphological closing(with a square structuring element of size 3 3 pixels).The vessel segmentation step is composed of 3 stages:highlighting of the vessels by application of a Black Top-Hat operator (BTH) (with a square structuringelement of size 31 31 pixels) (Serra, 1982),thresholding of the obtained image (the threshold valueis the mean gray value of the BTH image) and an arealtering to remove small components which are notconnected to the vascular tree. The skeletonization isthen applied to extract the medial axis of the vesselsby using the pattern thinning proposed by Zhang andSuen (1984). The obtained skeleton is then ltered

to extract undesirable branches and loops. The resultis a rened vessel skeleton which is nally used todetermine vessel endpoints.

2


3/6

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Fig. 3. Steps of vessel endpoints extraction: (a)FAF image, (b)gaussian ltering of (a), (c)application of a BTH operator on (b), (d)thresholding of (c), (e)areaopening of (d), (f)skeletonization of (e),(g)rened skeleton, (h)endpoints superposed on the rened skeleton.

Optic Disc detection In the studied FAF images,the OD appears as a dark oval region with a diameterof about 180 pixels. In the literature algorithms areoften divided into two main steps: the localizationand the OD boundary detection. As for the macula,the OD detection algorithms are image modalityspecic (more often performed on angiography andcolor images). In the color images, the high gray

level value and the elliptical shape features of theOD are exploited for its localization (Sinthanayothinet al. , 1999; Narasimha-Iyer et al. , 2006). In

(Siddalingaswamy and Prabhu, 2007), an optimaliterative threshold with connected component analysisis applied to identify the bright OD. In (Foracchiaet al. , 2004; Tobin et al. , 2007; Abdel-Razik Youssif et al. , 2008), the geometrical relationship between theOD and the vessels is used to localize the OD. In(Walter et al. , 2002), a watershed based segmentationtechnique using morphological ltering and shadecorrection is applied for the OD segmentation.The proposed algorithm for an automatic ODlocalization and detection is based on mathematicalmorphology operators (Serra, 1982), it is divided into5 steps (illustrated on Fig. 4): preprocessing, extractionof the large dark Connected Components (CCs) of size comparable to the one of the OD, selection of a candidate (from the set of the detected CCs by

analysis based on the endpoints of the blood vessels),extraction of the OD boundary by a marker-controlledwatershed and boundary smoothing, OD modeling byellipse tting in order to x its center.The preprocessing step is the same as the oneapplied for the preprocessing step of vessel endpointsextraction. The OD localization consists in detectingOD marker by application of a closing operator (witha square structuring element of size 41 41 pixels),which removes all thin dark details including vessels.The large components like the OD are preserved (Fig.4b).

For the extraction of large CCs, we propose an iterativethreshold segmentation method controlled by a squareshape constraint. In this method, the threshold valueis progressively lowered from the mean gray level.At each new thresholding step, the segmented CCssmaller than the maximum OD size are added tothe resulting image. The OD extraction steps areillustrated on the Fig. 4c.The next task is to select a candidate componentfrom the set of the CCs obtained at the end of theiterative process. As blood vessels originate from theOD, at each endpoint falling in the search window

around the CCs (1.5 times bigger than its boundingbox) is assigned a weight regarding the angle betweenthe vessel endpoint tangent and the CC center. TheCC presenting the maximum of vessel endpointscumulative weight is selected as a likely OD candidate(see Fig. 4d). The selected component is either a part of the OD or is slightly bigger than the OD in the presenceof pathologies around it.The OD boundary is extracted by a marker-imposedwatershed (Fig. 4e). The inner and outer markers aregenerated respectively by erosion and dilation of theselected CC. The extracted OD has an irregular shape,

caused by the similarity of the gray values between thevessels and the OD. No signicant edges exist betweenthe OD and the adjoining blood vessels.

3


4/6

C HAUDHRY A DNAN et al. : Automatic macula detection

In order to approximate the OD center, and to reducethe distortion in the extracted OD boundary, an ellipsetting is applied on the watershed image (Fig. 4f). Themethod used is based on a least square minimizationproposed by Halir and Flusser (1998). It guaranteesan ellipse-specic solution even for scattered or noisydata. Therefore, this algorithm overcomes the distortedboundary of OD and gives a smooth ellipse shape. Itgives also the opportunity to determine the center of the ellipse which could be associated as the one of theOD.

(a) (b)

(c) (d)

(e) (f)

Fig. 4. Steps of the OD extraction: (a)FAF image, (b)closing of (a), (c)Iterative thresholding of (b) under

shape constraint, (d) candidate selection procedure for OD, (e) marker imposed on the gradient image of (a),(f) ellipse tting and approximation of the OD center.

(a) (b)

(c)

Fig. 5. Steps of the macula detection and localizationapplied on the FAF image presented on Fig.3:(a)illustration of onion peeling algorithm, (b) result of BTH thresholding with the application of an area lter,(c) macular region.

Macula detection algorithm It consists of 3steps: a macular mask generation, a BTH by diameterclosing to exploit the hypouorescence of the macula

and a post processing step which uses together the twoprevious step results.Anatomically, the fovea is located in the temporaldirection with regard to the OD center, the distancebetween them is twice the OD diameter and there is anangle of about -15(resp 195) in the left eye (respright) between the horizontal line passing throughthe OD center and the line joining the OD centerto the fovea (Li and Opas, 2004; Siddalingaswamyand Prabhu, 2007). Furthermore, the macular zoneis located in the smallest convex hull of the centralavascular zone of the vessels. This convex hull can

be automatically determined from the blood vesselendpoints. From this convex hull, a mask could bedesigned to locate the likely region for the macularzone.The macular mask generation is performed asfollows(see Fig.5a): extraction of the vessel endpoints,positioning of a radial histogram (in relation to the ODcenter), application of the peeling onion algorithm.The onion peeling algorithm consists in removing theouter endpoints iteratively considering that radial bincontains at least one point.The BTH by diameter closing is a morphological

operator dened by equation 1 where o ( f ) is theclosing by diameter introduced by Walter (2003). For

the studied FAF images, parameter of the diameter

4


5/6


6/6

Ecs10_contrib

Documents

Transcript of Ecs10_contrib