ART-VENA: Retinal Vaseular Caliber Measurement

ART-VENA: Re inal Vascula Calibe Measu emen An onio Mosque a1, Raquel Dosil1, Víc o Lebo án1, Fe nando Pa do1, F ancisco Gómez-Ulla2,3, Bashi Hayik3, An onio Pose4 and Ma a Rod íguez4 1 G upo de Visión A i icial, Depa amen o de Elec ónica e Compu ación, Uni e sidade de San iago de Compos ela. Campus Uni e si a io Su s/n, 15706 San iago de Compos ela, Spain. [email p o ec ed], { dosil,nelbla, pa do}@usc.es 2 Depa amen o de Ci ugía, Uni e sidad de San iago de Compos ela. San F ancisco s/n, 15705 San iago de Compos ela, Spain. [email p o ec ed] 3 Se icio de O almología, Hospi al de Conxo, Complejo Hospi ala io Uni e si a io de San iago de Compos ela; Conxo, 15706 San iago de Compos ela, Spain. 4 Se icio de Medicina In e na, Hospi al de Conxo, Complejo Hospi ala io Uni e si a io de San iago de Compos ela; Conxo, 15706 San iago de Compos ela, Spain. Abs ac . The size o e inal ascula calibe in eye undus images is a unda- men al diagnosis pa ame e in he s udy o sys emic ascula pa hologies, like a e ial hype ension o a e ioscle osis. ART-VENA is a semiau oma ic sys em o measu e he e inal ascula calibe . F om he medical poin o iew, i s e- pea abili y (coe icien s o a ia ion unde 1.5%) u ns i in o a eliable ool o objec i y ascula changes which p e iously depended on he obse e ’s sub- jec i i y. 1 In oduc ion The au oma ic compu e ized analysis o blood essels om medical images has played, in he las yea s, an impo an ole in many clinical and esea ch s udies on a la ge numbe o diseases. The p esence o noise, he a iabili y o he ana omical undus and he low and changing con as o essels in many image modali ies make he eliable essel de ec ion a di icul ask. The e o e, au oma ic de ec ion becomes a key s a ing poin . Pa icula ly, e inal ascula calibe is a undamen al pa ame e in he s udy o sys emic ascula pa hologies, like a e ial hype ension o a e ioscle o- sis. Vessels p esen in he eye a e he i s o mani es he symp oms o he a e ial hype ension [1]. The abno mali ies in he e inal ascula iza ion can e lec he de- g ee o damage p o oked by hese diseases, an aspec ha decisi ely in luences he de elopmen o b ain and ca dio ascula complica ions. Be o e he analysis o he al e a ions in he e inal ascula calibe due o hese sys- emic diseases, i is necessa y o s udy he a ia ions caused by age in no mal pa ien s. This is a con o e sial opic in he medical ield. Se e al classic au ho s ha e ob- se ed ha old age is di ec ly ela ed o a e iola calibe al e a ions in no mal pa- ien s, hough he e is no an ag eemen on he causes o his ela ion [2, 3, 4]. On he con a y, o he au ho s do no ind e inal ascula calibe changes in old pa ien s [5, 6, 7]. I has o be aken in o accoun ha hese desc ip ions a e highly subjec i e. I would be e y use ul, he e o e, o de elop an e ec i e and ep oducible me hod o objec i ely e alua e he e inal ascula calibe . This objec i e measu e could de- ini i ely de e mine whe he obse ed al e a ions a e ela ed o age in no mal pa ien s, and, in case ha his ela ion exis s, i could dis inguish be ween hese changes and hose o igina ed by gene al diseases, like a e ial hype ension o a e ioscle osis. 1.1 Re inal Vascula Calibe Measu emen Re inal ascula calibe measu emen has been a widely s udied subjec in he a ea o Oph halmology. In he middle o he 20 h cen u y many wo ks begin o be ca ied ou wi h he aim o p edic ing, om hese measu es, he s a e and p ognosis o diseases like a e ial hype ension o a e ioscle osis. In a i s s age, om he 50’s o he 80’s, ascula wid h measu emen om eye undus pho og aphs was manually made by means o e inal image p ojec ions [8, 9, 10, 11, 12, 13]. Se e al wo ks p o e ha hese measu es can be pe o med wi h an e o in e io o 3 µ m a he e ina le el [11, 14]. Ob ained esul s u ned ou o be con adic o y in many cases, p obably as a esul o he high subjec i i y o he analy- sis and he lack o s a is ical s udies on he acqui ed da a. B inchmann-Hansen and col. ealized he i s s udies s a is ically analyzed [15]. They ound he calibe o h ee a e ies and h ee eins, achie ing a coe icien o a ia ion o 3.6 %. Pos e io s udies each a dec ease in he coe icien o a ia ion down o 2.3% [16, 17]. A he end o he 80’s decade, compu e ized essel measu emen echniques ap- pea ed. In he main, hese echniques use he compu e jus o isualize and s o e he da a. The spo ing o blood essel bounda y poin s is made manually by he specialis . Some wo ks eached coe icien s o a ia ion be ween 0.6% and 4.0%, wi h mean alue a ound 2.2% [18, 19, 20]. Especially impo an a e he wo ks o Polak and col. on he ep oducibili y and sensi i i y o he Zeiss’ analyze o e inal essels (Zeiss RVA) [21]. In size measu emen s o a e ies and eins hey ob ained coe icien s o a ia ion alues o 1.3% and 2.6% espec i ely. These da a place he Zeiss’ essel analyze among he e e ence comme cial ins umen s in his a ea. 1.2 Vessel De ec ion Au oma ic de ec ion o essels has in ol ed much esea ch in he ield o digi al im- age p ocessing du ing he las yea s. Many wo ks in he li e a u e add ess he p oblem o au oma ic de ec ion o blood essels in a ious domains. T acking o essels has been one o he mo e s udied echniques in ecen yea s [22, 23]. In gene al, gi en a s a ing essel con ou poin and one sea ch di ec ion, essel acking in ol es explo ing he image o ind o he con ou poin s. The c i e- ion o de ine bounda y poin s is based on he analysis o he pixel g ay le el alues on a sec ion ha is o hogonal o he acking di ec ion. Mo e sophis ica ed e sions o his idea ans o m he p oblem o essel con ou de ec ion in ha o g aph op imal pa h sea ch [24]. O he a i icial in elligence o compu e ision me hods ha e been also p oposed o ob ain a comple e au oma ion o he de ec ion p ocedu e, exploi ing in a mo e exhaus i e way he a ailable ana omical knowledge [25, 26]. The majo i y o he de eloped me hods adop bounda y de ec ion as he basic ool o essel de ec ion. Since essel con ou s a e usually smoo h and uzzy and he un- dus is noisy, a numbe o au ho s ha e applied, as an ini ial s a egy, he maximiza ion o essel enhancemen . Some app oaches a e based on o a ion in a ian ope a o s o linea mo phological ope a o s o sui able size and o ien a ion, which a e applied o he image o minimize noise [25, 27]. Al e na i e app oaches in ol e image con olu- ion wi h mul iple il e s pe model, each o hem designed o de ec essels o gi en size and o ien a ion [28]. The main d awback o hese me hods is hei high compu a- ional cos , which ende s hem p ac ically unusable o on-line diagnosis assis ance sys ems o sys ems equi ing ela i ely sho esponse imes. Zana and Klein de el- oped a essel bi u ca ion de ec ion me hod as a p e ious s ep o image egis a ion [29]. Quek and Ki bas ca ied ou a me hod o essel ex ac ion om angiog aphies based on wa e p opaga ion [30]. 2 Re inal Vascula Calibe Measu emen The measu emen me hod de eloped in his wo k is nea ly comple ely au oma ic. The ope a o is ini ially equi ed jus o ma k he op ic ne e, hough high accu acy is no necessa y. Once his ini ial e e ence is se , he au oma ic sequence o de ec ion and measu emen o he essels on he image s a s. The me hod ou pu is a lis o he s uc u es iden i ied as essels and hei wid hs, eady o be analyzed by he expe . As he undus ascula s uc u es a e inhe en ly adial, he wid h measu emen o e inal essels will be pe o med o e a ious ci cum e ences cen e ed on he op ic ne e, a dis ances ha a e mul iple o he op ic disk adius, depic ed by he ope a o (a common p ocedu e in Oph halmology). Using a pola coo dina e space, a linea ep esen a ion o he g ay le el p o iles is ob ained o each o he analyzed ci cum- e ences, as shown in le side o Fig. 1. The g ay le el p o ile along each ci cum e - ence is cha ac e ized by he p esence o alleys, which a e po en ial essels. They a e placed o e he noise o he ana omical undus ha is nea ly uni o mly dis ibu ed. Be o e i s de ec ion, a p ep ocessing s age is needed o minimize noise e ec s. The employed il e o noise educ ion mus no modi y he loca ion o he p o ile ansi ion egions, which de e mine essel wid hs in he measu emen phase. To mini- mize noise, a median il e has been chosen. I pe mi s o emo e g ay le el alues ha di e oo much om he su ounding alues. This il e ing educes he de ec ion o alleys caused by noise pe u ba ions induced by he ana omical undus. The me- dian il e used in his wo k has a window size o 11 pixels cen e ed in he pixel unde analysis. One impo an p ope y o his il e is ha i does no al e he magni ude o he g ay le el g adien and, he e o e, espec s he posi ion o ansi ions. The igh side o Fig. 1 shows he p o ile ob ained a e he il e ing p ocess. Al hough he median il e elimina es many noise alleys, some o hem, p oduced by illumina ion a ia ions and he p esence o undus ana omical s uc u es o he han essels, s ill emain, oge he wi h he blood essel alleys. Fo ha eason, he simple de ec ion o il e ed sequence minima does no wa an y co ec iden i ica ion o es- sels. The de ini i e de ec ion o blood essels alleys is es ablished as a unc ion o hei 1D and 2D mo phology. In his wo k he hypo hesis ha he a e sal essel p o iles a e deep is assumed (1D model), i.e., hei g ay le els a e much smalle han he su ounding undus ones. Consequen ly, he in ensi y di e ences be ween essel alleys and undus a e g ea e han hose o noise alleys. Mo eo e , he essel alley s uc u e is s able in he adial di ec ion ega ding he op ic disk (2D model). Fig. 2 shows he 1D essel model. Poin s x1, x2 and x3 co espond espec i ely o he si es o he p e ious maximum, he minimum and he pos e io maximum on he alley s uc u e. Pa ame e s y1, y2 and y3 a e he p o ile heigh s (g ay le els) measu ed a x1, x2 and x3 espec i ely. Using he h ee heigh pa ame e s, a alley quali y meas- u e e lec ing ela i e dep h is es ablished. This measu e is gi en by he exp ession 31 2 2 1yy y Q + −= . (1) Valleys wi h Q alues g ea e han 0.06 a e conside ed candida es o o m blood essels. This p ocess is ealized indi idually o e 2n+1 ci cum e ences, wi h n posi- i e in ege and wi h hei adius di e ing in one pixel, being he inne mos ci cum e - ence adius selec ed by he specialis . The n alue, which can be modi ied by he ope a o , is ecommended o be se be ween 5 and 9. Smalle alues do no wa an y he s aigh essel hypo hesis which will be applied in he wid h measu emen s age. These da a a e he inpu o he 2D modeling phase. Fo a egion o be conside ed a ue essel, a alley mus appea in 2/3 o he analyzed ci cum e ences. In addi- ion, he angula dis ance be ween alleys asso- cia ed o consecu i e ci cum e ences mus be lesse han 0.089 adians. In he op le hand side o Fig. 3, a sequence o 11 alleys (n=5) can be obse ed, which e i ies he 1D and 2D imposed condi ions, so ha hey de e mine he exis ence o a blood. Fig. 1. Le : G ay le el p o ile along one ci cum e ence cen e ed a he op ic ne e. Righ : G ay le el p o ile a e median il e applica ion. Fig. 2. Vessel model When he essel posi ions a e loca ed, he nex s ep is wid h measu emen . This ask is accomplished by loca ing he ex emes o he g ay le el de i a i e along he angula di ec ion, i.e., along he p o iles. Hence, posi ions o essel walls a e se a he maximum ansi ion poin s o he in ensi y p o ile. These ex eme poin s a e ob ained by means o quad a ic in e pola ion ( i o a pa abola) o he de i a i e alues, o ob ain subpixel accu acy. In his way, po en ial image quan iza ion ambigui ies a e a oided and esul s a e be e assessed. Assuming he hypo hesis o he analyzed es- sel segmen is s aigh , he a ailable poin s can be i ed o a s aigh line, as shown on he bo om le hand side and on he igh hand side images in Fig. 3. Calibe is inally es ima ed o e a line pe pendicula o he bisec o o he lines i ed o bo h essel bo de s, as can be seen in Fig. 3. 3 Resul s In o de o alida e he de eloped p ocedu e, a ep oducibili y/ eliabili y s udy has been designed in h ee s ages. These phases a e in ended o e alua e he ep oducibil- i y/ eliabili y o he me hod agains di e se ac o s ha can in luence he s udy. These ac o s include a iabili y on he measu emen s made by an obse e , a iabili y among di e en obse e s and a iabili y due o he pa ien unde s udy. Once he goals o his expe ience we e explained o he olun ee s, chosen om he companion o he pa ien s ha came o he Se icio de U gencias de O almología o he Complejo Hospi ala io Uni e si a io de San iago de Compos ela (Spain) and a e hey had signed he in o med consen o p oceed wi h pupil dila ion and an oph hal- mologic exam, he olun ee pupil was dila ed using an ins illa ion o wo d ips o opicamide and wo d ips o phenyleph ine. A e a pe iod o 20 o 30 minu es, an eye undus pho og aph was aken om bo h eyes o each subjec a 50 deg ees, wi h ane i a ligh (540 nanome e s) and cen e ed a he papilla. The came a used o ake he undus pho og aphs was he Topcon TRC-50 IA connec ed o he IMAGEne 1024 Digi al Imaging Sys ems by Topcon Ins umen s, Pa amus, NJ. All cap u ed images Fig. 3. Le : Sequence o p o iles e i ying 1D and 2D modeling condi ions. Righ : Geome ic calculus o he s aigh line ha is used o de e mine de he blood essel wid h. we e di ec ly s o ed in op ical disks and displayed on a high esolu ion moni o (1280x1024 pixels) o hei isualiza ion and pos e io analysis. The di ec digi aliza- ion o he cap u ed image was made by means o a digi al came a connec ed o he a o emen ioned undus came a. The elimina ion o he in e media e s ep, usually in- cluded in many o he p ocedu es, a oids he in oduc ion o e o ac o s [31]. To achie e he objec i es o his wo k, a basic popula ion sample was o med by 120 no mo ensi e olun ee s (240 undus images), wi hou any known ascula pa- hologies. A subjec is classi ied as no mo ensi e in he absence o a p e ious diagno- sis o a e ial hype ension and he p esence o PAS/PAD quan i ies unde 140/90, ollowing he p ocedu es o he Ame ican Na ional Commi ee o HBP [32]. Subjec ages anged om 10 o 69 yea s, wi h a 1:1 a io be ween sexes. Each age g oup (10- 19; 20-29; 30-39; 40-49; 50-59; 60-69) is cons i u ed by 20 indi iduals, 10 males and 10 emales. In all cases he olun ee s had anspa en media in bo h eyes and a e ac- i e de ec in e io o 3 diop e s o hype me opia, myopia and as igma ism. Table 1 shows he esul s ob ained o he coe icien o a ia ion o e he indi id- ual measu es and he Pea son’s momen um co ela ion coe icien (p<0.001) o e he global measu es. The in aobse e ep oducibili y indica es ag eemen in measu e- men s made by he same obse e when epea ing he measu e o e he same se o images. To analyze his ac o , one single obse e e alua ed 40 images co esponding o 40 eyes o 20 indi iduals (20 le eyes and 20 igh eyes) wo imes, sepa a ed om each o he by an in e al o 24 hou s, wi hou knowing no he esul s nei he he measu emen poin s in he p e ious a emp . The in e obse e ep oducibili y indi- ca es ag eemen in measu emen s ealized independen ly by wo di e en obse e s. In his case, wo obse e s e alua ed, wi hou in o ma ion exchange, 30 images co e- sponding o 30 eyes o 15 indi iduals (15 le eyes and 15 igh eyes). In aindi idual ep oducibili y indica es conco dance in measu emen s, ealized in a blind ashion, by he same obse e o e images o he same subjec and same eye aken wi h a di e - ence in some ime in e al. To analyze his ac o , one single obse e e alua ed wo images o he same eye, sepa a ed by a ime pe iod o one mon h, epea ing he p oc- ess o 30 eyes o 15 indi iduals (15 le eyes and 15 igh eyes), wi hou knowing no he esul s nei he he measu emen poin s in he p e ious ial. Table 1. Coe icien o a ia ion (CV) and Pea son’s momen um co ela ion coe icien ( ) ob ained in he a ious plan ed expe imen s. CV ( ) C In aobse e In e obse e In aindi idual Polak, 2000 A e y 1.10 % (0.96) 1.30 % (0.94) 1.08 % (0.93) 1.3 % Vein 0.98 % (0.91) 1.47 % (0.86) 1.12 % (0.93) 2.6 % A e io enous index 0.50 % (0.99) 0.99 % (0.95) 0.40 % (0.99) 4 Discussion Rep oducibili y o measu emen me hods is a c ucial aspec in Medicine [31]. Fo his eason, he i s plan ed goal in his wo k is o demons a e ha epea ed ascula calibe measu emen s o e he same pho og aph wi hou knowing he e e ence poin s b ing he same esul s. As a second objec i e, i has been p oposed o demons a e ha he same hing happens when one ce ain pho og aph is measu ed by wo di e en obse e s. And, as a hi d objec i e, o demons a e also ha ascula calibe meas- u emen s om wo pho og aphs o he same indi idual wi h a ime di e ence o one mon h do no p esen signi ican a ia ions. In aobse e , in e obse e and in aindi idual a iabili ies a e small in ou me hod and he measu emen ep oducibili y in he h ee g oups p esen s a e y high deg ee o conco dance, wi h a Pea son’s momen um co ela ion coe icien (equi a- len o he in aclass co ela ion coe icien ) o e 0.86 and a coe icien o a ia ion in e io o 1.47% in all cases, while Polak e col. ob ained coe icien o a ia ion alues be ween 1.3% and 2.6% in hei wo k wi h he Zeiss’ e inal essel analyze (Zeiss RVA) [21]. In pa icula , his coe icien o a ia ion supe io bound is he bes ob ained by he whole o he p e iously commen ed me hods. These esul s ende he p esen ed me hod a secu e ool o objec i y mic o ascula changes which p e iously depended on he obse e ’s subjec i i y. The esea ch p ojec “Desa ollo y Validación de un Mé odo de Medida del Calib e Mic o ascula Re iniano en la Población No mal” ca ied ou using ART-VENA (insc ibed in he Spanish IP Regis e , wi h numbe 2001/15/934) was p ized wi h he P emio de In es igación Academia Médico-Qui ú gica /Conselle ía de Sanidade o he Xun a de Galicia (Spain). Re e ences 1. EYE ac s, “Sys emic Disease and he Eye”, UIC Eye Cen e , Uni e si y o Illinois a Chi- cago (2002) 2. Ballan yne, A.J., T ans. Oph halmol. Soc. UK., 57, 301-318 (1937) 3. Vogelius, H. and Bechgaa d, P., “The Oph halmoscopical Appea ance o he Fundus Oculi in Elde ly Pe sons wi h A e ioscle osis and No mal Blood P essu e”, B . J. Oph halmol., 34, 404-408 (1950) 4. Leishman, R., “The Eye in Gene al Vascula Disease: Hype ension and A e ioscle osis”, B . J. Oph halmol., 41, 641-701 (1957) 5. Kagan, A., Au ell, E. and Dob ee, J.A., “A No e o Signs in he Fundus Oculi and A e ial Hype ension Con en ional Assessmen and Signi icance”, Bull. Wo ld Heal h O gan., 34, 955-960 (1966) 6. Aoki, N., Ho ibe, H. and col., “Epidemiological E alua ion o Funduscopic Findings in Ce eb o ascula Diseases III: Obse e Va iabili y and Rep oducibili y o Funduscopic Findings”, Jpn. Ci c. J., 41, 11-17 (1977) 7. Dimmi , S.B., Wes , J.N. and col., “Use ulness o oph halmoscopy in mild o mode a e hype ension”, Lance , 1, 1103-1106 (1989) 8. Boyd, T.A.S. and de Ma ge ie, J., “Calibe o Re inal A e ioles in Hype ension”, T ans. Can. Oph halmol. Soc., 23, 65-76 (1960) 9. Hill, D.W. and Dolle y, C.T., “Calibe Changes in Re inal A e ioles”, T ans. Oph hal. Soc. UK., 83, 61-70 (1963) 10.Michaelson, I.C., The Ocula Ci cula ion in Heal h and Disease, Can. J. S., Ed. Kimp on, London (1969) 11.Hodge, J.V., Pa , J.C. and Spea s, S., “Compa ison o Me hods o Measu ing Vessel Wid hs on Re inal Pho og aphs and he E ec o Fluo escein Injec ion on Appa en Re inal Vessel Calibe s”, Am. J. Oph halmol., 68, 1060-1068 (1969) 12.Sanchéz-Salo io, M., Re inopa ías Vascula es., Ed. G a insa (1971) 13.Pa , J.C. and Spea s, G.F.S., “Gene al Calibe o he Re inal A e ies Exp essed as he Equi alen Wid h o he Cen al Re inal A e y”, Am. J. Oph halmol., 77, 472-477 (1974) 14.Pa , J. C., “Hype ensi e Gene alized Na owing o Re inal A e ies”, T ans. Oph halmol. Soc. NZ, 26, 55-60 (1974) 15.B inchmann-Hansen, O. and Heie , H., “The Appa en and he T ue Wid h o he Blood Column in Re inal Vessels”¸ Ac a Oph halmologica, Supplemen 179. 64, 29-32 (1986) 16.Delo i, F.C., Fi ch, K.A. and col., “E alua ion o Mic ome ic and Mic odensi ome ic Me hods o Measu ing he Wid h o Re inal Vessels Images on Fundus Pho og aphs”, G ae e’s A chi e o Clinical and Expe imen al Oph halmology, 226, 393-399 (1988) 17.S an on, A.V., Mullaney, P. and col., “A Me hod o Quan i ying Re inal Mic o ascula Al e a ions Associa ed wi h Blood P essu e and Age”, J. Hype ens., 13, 41-48 (1995) 18.Meehan, R.T., Taylo , G.R. and col., “An Au oma ed Me hod o Quan i ying Re inal Vascu- la Response du ing Exposu e o No el En i onmen al Condi ions”, Oph halmology, 97, 875-881 (1990) 19.Rassam, S.M.B., Pa el, V. and Col, “Accu a e Vessel Wid h Measu emen om Fundus Pho og aphs: A New Concep ”, B . J. o Oph halmology, 78, 24-29 (1994) 20.Pede sen, L., G unkin, M. and col., “Quan i a i e Measu emen o Changes in Re inal Vessel Diame e in Ocula Fundus Images”, Pa e n Recogni . Le ., 21, 1215-1223 (2000) 21.Polak, K., Do ne , G. and col., “E alua ion o he Zeiss Re inal Vessel Analyse ”, B . J. Oph halmol., 84, 1285-1290 (2000) 22.Liu, I. and Sun, Y., “Recu si e T acking o Vascula Ne wo ks in Angiog ams based on he De ec ion-Dele ion Scheme”, IEEE T ans. Med. Imaging, 12, 2, 334-341 (1993) 23.Miles, F.P. and Nu all, A.L., “Ma ched Fil e Es ima ion o Se ial Blood Vessels Diame- e s om Video Images”, IEEE T ans. Med. Imaging, 12, 2, 147-152 (1993) 24.Sonka, M., Wilb ich , C.J. and col., “Simul aneous De ec ion o bo h Co ona y Bo de s”, IEEE T ans. Med. Imaging, 12, 3, 588-599 (1993) 25.Xia, W. and Lü, W., “Co espondence Analysis o Regional T acking in Co ona y A e- iog ams”, IEEE T ans. Med. Imaging, 11, 2, 153-160 (1992) 26.Coppini, M., Demi, M. and col., “An A i icial Vision Sys em o X- ay Images o Human Co ona y T ees”, IEEE T ans. Pa e n Anal. Mach. In ell., 15, 2, 156-162 (1993) 27.Thack ay, B.D. and Nelson, A.C., “Semi-au oma ic Segmen a ion o Vascula Ne wo k Images using a Ro a ion S uc u ing Elemen (ROSE) wi h Ma hema ical Mo phology and Dual Fea u e Th esholding”, IEEE T ans. Med. Imaging, 12, 3, 835-392 (1993) 28.Ge ig, G., Kolle , G. Th. and col., “Segmen a ion and Symbolic Desc ip ion o Ce eb al Vessel S uc u e Based on MR Angiog aphy Volume Da a”, Compu e Assis ed Radiology CAR`93, Lemke, H. U., Inamu a, K., Ja e, C.C. and Félix, R., edi o s, 359-365 (1993) 29.Zana, F. and Klein, J.C., “A Mul imodal Regis a ion Algo i hm o Eye Fundus Images using Vessels De ec ion and Hough T ans o m”, IEEE T ans. Med. Imaging, 18, 5, 419-428 (2001) 30.Quek, F.K.H. and Ki bas, C., “Vessel Ex ac ion in Medical Images by Wa e-P opaga ion and T aceback”, IEEE T ans. Med. Imaging, 20, 2, 117-131 (2001) 31.Coupe , D.J., Klein, R. and col., “Reliabili y o Re inal Pho og aphy in he Assessmen o Re inal Mic o ascula Cha ac e is ics: The A e ioscle osis Risk in Communi ies S udy”, Am. J. Oph halmol., 133, 78-88 (2002) 32.Join Na ional Commi ee on he P e en ion, De ec ion, E alua ion, and T ea men o High Blood P essu e, Six h Repo A ch. In e n. Med., 157, 21, 2413-2446 (1997)