Comparison of visual assessment and computer image analysis of intracoronary thrombus type by optical coherence tomography

RESEARCH ARTICLE Compa ison o isual assessmen and compu e image analysis o in aco ona y h ombus ype by op ical cohe ence omog aphy Timo P. Kai osoja 1 , Shengnan Liu 2 , Jouke Dijks a 2 , Heini Huh ala 3 , Tej She h 4 , Olli A. Kajande ID 1 * 1Hea Hospi al, Tampe e Uni e si y Hospi al and Facul y o Medicine and Li e Sciences, Uni e si y o Tampe e, Tampe e, Finland, 2Di ision o Image P ocessing, Depa men o Radiology, Leiden Uni e si y Medical Cen e , Leiden, Ne he lands, 3Facul y o Social Sciences, Uni e si y o Tampe e, Tampe e, Finland, 4McMas e Uni e si y and Popula ion Heal h Resea ch Ins i u e, Hamil on Heal h Sciences, Hamil on, Canada *olli.kajande @sydansai aala. i Abs ac Backg ound Analysis o in aco ona y h ombus ype by op ical cohe ence omog aphy (OCT) imaging is highly subjec i e. We aimed o compa e a newly de eloped image analysis me hod o sub- jec i e isual classi ica ion o h ombus ype iden i ied by OCT. Me hods Thi y pa ien s wi h acu e ST ele a ion myoca dial in a c ion we e included. Th ombus ype isually classi ied by wo independen eade s was compa ed wi h analysis using QCU- CMS so wa e. Resul s Repea abili y o he compu e -based measu emen s was good. By using a ROC, a ea unde cu e alues o disc imina ion o whi e and ed h ombi we e 0.92 (95% con idence in e als (CI) 0.83–1.00) o median a enua ion, 0.96 (95% CI 0.89–1.00) o mean backsca e and 0.96 (95% CI 0.89–1.00) o mean g ayscale in ensi y. Median a enua ion o 0.57 mm -1 (sensi i i y 100%, speci ici y 71%), mean backsca e o 5.35 (sensi i i y 92%, speci ici y 94%) and mean g ayscale in ensi y o 120.1 (sensi i i y 85%, speci ici y 100%) we e iden i- ied as he bes cu -o alues o di e en ia e be ween ed and whi e h ombi. Conclusions A enua ion, backsca e and g ayscale in ensi y o h ombi in OCT images di e en ia ed ed and whi e h ombi wi h high sensi i i y and speci ici y. Measu emen o hese con inuous pa ame e s can be used as a less use -dependen me hod o cha ac e ize in i o h ombi. The clinical signi icance o hese indings needs o be es ed in u he s udies. PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 1 / 15 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Ci a ion: Kai osoja TP, Liu S, Dijks a J, Huh ala H, She h T, Kajande OA (2018) Compa ison o isual assessmen and compu e image analysis o in aco ona y h ombus ype by op ical cohe ence omog aphy. PLoS ONE 13(12): e0209110. h ps:// doi.o g/10.1371/jou nal.pone.0209110 Edi o : Elisabe a Rico ini, Campus Bio-Medico Uni e si y o Rome, ITALY Recei ed: Augus 21, 2018 Accep ed: No embe 28, 2018 Published: Decembe 17, 2018 Copy igh : ©2018 Kai osoja e al. This is an open access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion License, which pe mi s un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal au ho and sou ce a e c edi ed. Da a A ailabili y S a emen : All ele an da a a e wi hin he manusc ip and i s Suppo ing In o ma ion iles. Funding: OAK ecei ed academic esea ch unding om The compe i i e esea ch und o Pi kanmaa Hospi al Dis ic , Finland, www. ays. i/en-US. The unde had no ole in s udy design, da a collec ion and analysis, decision o publish, o p epa a ion o he manusc ip . Compe ing in e es s: The au ho s ha e decla ed ha no compe ing in e es s exis . In oduc ion In aco ona y h ombus is a equen inding in pa ien s wi h acu e co ona y synd omes unde going in asi e angiog aphy and pe cu aneous co ona y in e en ion (PCI). Resea ch on he clinical signi icance o h ombus ype is ongoing, and wi h mo e de ailed unde s anding o he h ombo ic p ocess, he goal is o acili a e he de elopmen o ailo ed he apies o pa ien s wi h acu e co ona y synd omes. Op ical cohe ence omog aphy (OCT) is a high- eso- lu ion in a ascula imaging echnique wi h excellen con as be ween he essel lumen and in a ascula s uc u es [1]. In a consensus pape , he e idence le el o image in e p e a ion o in aco ona y h ombus using OCT has been conside ed o be high [2]. The abili y o OCT o p oduce images allowing di e en ia ion be ween h ombus ypes has been alida ed agains his ology in pos -mo em samples o human co ona y a e ies [3]. I is widely accep ed in he OCT communi y ha h ombi in OCT images can be classi ied as high-backsca e ing wi h a signal- ee shadowing (‘ ed’, e y h ocy e- ich) o low-backsca e ing (‘whi e’, h ombocy e- ich) [2]. This encou ages u he alida ion in clinical pa ien s. Howe e , alid analysis o di - e en h ombus ypes in i o by OCT is hinde ed by he high deg ee o subjec i i y o mea- su emen and he lack o alida ed me hodology. The aim o he p esen s udy is o de elop a compu e image analysis -based me hod o he assessmen o h ombus mo phology in OCT images ha can be used o u he in i o ali- da ion. Compu e algo i hms de e mining backsca e , a enua ion and in ensi y o he OCT signal a e es ed and compa ed o he p esen s anda d me hod, which is consensus classi ica- ion o h ombus ype by wo independen analys s. Ou hypo hesis is ha applying compu e image analysis o assess h ombus ype is easible and can educe obse e a ia ion in com- pa ison o subjec i e e alua ion. Pa ien s and me hods Pa ien s The s udy was app o ed by he E hical boa d a Pi kanmaa Hospi al Dis ic , app o al num- be R10131. W i en in o med consen was ob ained om all pa icipan s. The s udied se ies included 30 pa ien s en olled in an OCT subs udy o he TOTAL ial. The TOTAL ial was an in e na ional, mul icen e, andomized ial o ou ine h ombec omy (using he Expo ca he e , Med onic Ca dio ascula , San a Rosa, CA, USA) compa ed wi h PCI alone in STEMI pa ien s ea ed wi h PCI (n = 10732) [4,5]. A p ospec i e OCT sub-s udy (n = 214) o TOTAL ial e alua ed h ombus bu den in pa ien s wi h symp oms o myoca dial ischae- mia las ing o �30 min and de ini e elec oca diog aphic changes indica ing STEMI who we e e e ed o p ima y PCI and andomized wi hin 12 h o symp om onse [6]. Pa icipa- ion in he s udy equi ed es o a ion o TIMI 2–3 low a e he i s de ice. Pa ien s we e excluded om he OCT sub-s udy i hey we e in ca diogenic shock o had known enal ail- u e. In o med consen was ob ained om all indi idual pa icipan s included in he s udy. O he pa ien s in he OCT subs udy who we e andomized o h ombec omy, 72 had good qual- i y OCT pullbacks a ailable p eceding balloon p edila a ion. Ou o hese, 30 pa ien s wi h he la ges maximal h ombus a eas we e selec ed o he p esen s udy ( o de ailed pa ien low-cha , see S1 Fig). All p ocedu es pe o med in s udies in ol ing human pa icipan s we e in acco dance wi h he e hical s anda ds o he ins i u ional and/o na ional esea ch commi ee and wi h he 1964 Helsinki decla a ion and i s la e amendmen s o compa able e hical s anda ds. This a icle does no con ain any s udies wi h animals pe o med by any o he au ho s. Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 2 / 15 OCT imaging OCT imaging was pe o med using he Ilumien OCT sys em and C7 D agon ly ca he e (S Jude Medical, Minneso a, MN, USA). The adio-opaque dis al ma ke o he OCT ca he e was posi ioned 1–2 cm dis al o he a ge lesion o s en and con as was injec ed ei he man- ually o by au oma ic injec o , acco ding o local p ac ice. Adequa e image quali y was checked and addi ional images we e ob ained, i he image quali y was sub-op imal. OCT imaging aw da a was expo ed in digi al o ma o o -line analysis. Op ical cohe ence omog aphy image analysis Da a on h ombus c oss-sec ional a eas a he culp i lesion was ob ained om p e ious analy- sis (6). Fo each pa ien , h ee OCT c oss-sec ions wi h he la ges h ombus a eas we e analyzed. Visual assessmen o h ombus ype Visual h ombus cha ac e is ics we e e alua ed by wo independen eade s (T.P.K. and O.A.K.) o assess he in e -obse e ep oducibili y. One obse e (T.P.K.) epea ed he analy- sis a e >4 week pe iod o assess he in a-obse e ep oducibili y. Th ombus was g aded by a six-s age scale, whe e alue 1 ep esen s pu ely whi e h ombus and alue 6 pu ely ed h ombus, c ea ing h ombus a enua ion sco e (TAS). In acco dance o published s anda ds [1,2], h ombus was conside ed whi e o mos ly whi e, when he abluminal bo de o he h ombus o he essel wall behind i could be seen. Con e sely, he mo e he h ombus caused shadowing and poo isibili y o he s uc u es behind i , he edde i was sco ed. In addi ion, we applied only wo addi ional ules. I a c oss-sec ion con ained mul iple agmen s wi h di - e en isual appea ances, TAS was de ined o he la ges agmen . In he case o mul iple same-sized h ombi, TAS was de ined o he mos s ongly a enua ing pa o he h ombus. Compu e image analysis o h ombus A special e sion o QCU-CMS e sion 4.69 so wa e (Leiden Uni e si y Medical Cen e , Lei- den, The Ne he lands) was used by wo independen analys s. Fi s , he h ombus a eas we e manually aced. The so wa e hen de e mined a enua ion coe icien , backsca e e m and g ayscale in ensi y alues o hese a eas and calcula ed se e al s a is ical pa ame e s, including median, mean, s anda d de ia ion and 5 h , 10 h , 90 h and 95 h pe cen ile alues. Two pa ien cases wi h whi e and ed h ombus showing he me hod a e shown in Fig 1. To a oid he inclu- sion o noise pixels and da k a eas (low a enua ion) a he abluminal side o he h ombus, we expe imen ed wi h di e en minimum h eshold alues o a enua ion. Fo he p esen analy- ses, a enua ion was analyzed using bo h 0.15 mm -1 and 0.5 mm -1 h eshold alues. The ligh a enua ion coe icien and backsca e e m ha e been de e mined based on a “dep h- esol ed” model [7]. The a enua ion coe icien desc ibes he o al dec easing a e o ligh when i is a eling hough he issue. OCT images a e gene a ed by ecei ing he back- sca e ed ligh . The backsca e coe icien desc ibes he e iciency o issue sca e ing ligh backwa ds. The backsca e e m is es ima ed o be ela ed o he backsca e coe icien . Mo e in o ma ion abou he es ima ion can be e e ed o he echnical pape [7]. The ad an age o his me hod is i s as pixel-wise es ima ion, which equi es no p ede ined delinea ion and is mo e lexible o pos analysis. Fo each pa ien , a weigh ed a e age o each s a is ical pa ame e was calcula ed using he pixel coun in each aced h ombus a ea. To s udy in e -obse e ep oducibili y, wo eade s analyzed he same OCT c oss-sec ions o 10 andomly selec ed pa ien s independen ly. Fo Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 3 / 15 in a-obse e compa ison, one obse e epea ed he analyses on hese 10 pa ien s a e a 4- week pe iod. Fo he epea ed assessmen s, OCT aw da a was e-impo ed in o he QCU- CMS so wa e. Then he h ombus a eas we e manually e- aced in he same ans e sal ames. S a is ical analysis Bland–Al man (B-A) analysis was used o es ima e he bias esul ing om in e - and in a- obse e a ia ion [8]. In he analysis, he mean alues o obse e 1 and 2 we e plo ed agains he di e ence o measu emen s be ween he obse e s. The 95% limi s o ag eemen (equal o ±1.96 SD) we e de e mined o he a iables. Mean ela i e di e ence o each a iable was de ined as he mean absolu e di e ence di ided by mean o obse e s 1 and 2. In aclass co e- la ion coe icien s (ICCs) and hei 95% con idence in e als we e calcula ed based on abso- lu e-ag eemen , 2-way mixed e ec model. Spea man co ela ions we e calcula ed be ween he con inuous a iables and TAS. In e - and in aobse e eliabili y o h ombus classi ica ion was assessed using he kappa s a is ic (κ). Recei e -ope a ing cha ac e is ic (ROC) cu e anal- yses we e pe o med o de e mine he bes cu o alues (using highes Youden Index) o QCU-CMS pa ame e s o disc imina ing ed and whi e h ombi and he a eas unde he cu e (AUCs), as well as he sensi i i ies and speci ici ies o he diagnos ic es . In addi ion, bina y logis ic eg ession analysis was used o e alua e how he di e en con inuous image analysis so wa e pa ame e s p edic ed he ype o h ombus. S a is ical analyses we e pe - o med wi h he S a a S a is ical so wa e: Release 13 (S a aCo p, College S a ion, TX, USA) and he IBM SPSS 21 S a is ics so wa e (IBM, A monk, NY, USA). Fig 1. Examples o h ombus assessmen by he QCU-CMS image analysis so wa e in OCT image ames o wo pa ien s wi h whi e (A) and ed h ombus (D). A e manual acing o he h ombi (B,E), a enua ion analysis was pe o med including egions wi h a enua ion alues abo e he designa ed h eshold alue (displayed in blue) (C,F). Accu a e segmen a ion o he luminal bo de o he h ombus and he con as - illed low a ea o he essel can be seen in C and F. �) OCT ca he e , #) guidewi e a e ac , ¤) h ombus, §) essel wall. OCT, op ical cohe ence omog aphy. h ps://doi.o g/10.1371/jou nal.pone.0209110.g001 Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 4 / 15 Resul s Baseline pa ien cha ac e is ics a e shown in Table 1. Quali a i e ( isual) h ombus classi ica ion by wo obse e s Fo he pu pose o s a is ical compa ison o he h ombus assessmen me hods, a six-s age scale consensus TAS was used o o m a bina y classi ica ion, whe e sco es 1–3 equaled whi e and 4–6 ed h ombus. The e we e 13 cases classi ied as whi e and 17 as ed h ombus (S2 Fig). Kappa alues o bina y classi ica ion o he h ombi we e 0.74 and 0.53, espec i ely, o in e - and in aobse e compa isons. Rep oducibili y o quan i a i e measu emen s using QCU-CMS so wa e Resul s p esen ed in Tables 2and 3show in e - and in aobse e a iabili y o measu e- men s in h ombus egions. Mean di e ences o mos s a is ical pa ame e s o h ombus Table 1. Baseline pa ien cha ac e is ics. N = 30 Age, yea s (mean ±SD) 57 ±10.1 Male sex, N (%) 22 (73.3) Risk ac o p o ile, N (%) Hype ension 14 (46.6) Diabe es 4 (13.3) Cu en smoking 8 (26.7) Obesi y (BMI �30) 14 (46.7) P io MI 2 (6.7) P io PCI 1 (3.3) C ea inine, μmol/L (mean ±SD) 82.0 ±16.6 An i h ombo ic ea men , N (%) Up on glycop o ein inhibi o IIb/IIIa use 11 (36.7) UFH p io o p ocedu e 22 (73.3) Enoxapa in p io o p ocedu e 2 (6.7) Bi ali udin p io o p ocedu e 0 (0.0) Culp i essel, N (%) RCA 16 (53.3) LAD 12 (40.0) LCx 2 (6.7) TIMI low�, N (%) 0 15 (50.0) 1 4 (13.3) 2 3 (10.0) 3 8 (26.7) TIMI h ombus class�, N (%) <3 4 (13.3) �3 26 (86.7) �p io o h ombec omy. BMI, body mass index; LAD, le an e io descending co ona y a e y; LCx, le ci cum lex co ona y a e y; MI, myoca dial in a c ion; PCI, pe cu aneous co ona y in e en ion; RCA, igh co ona y a e y; SD, s anda d de ia ion; TIMI, h ombolysis in myoca dial in a c ion; UFH, un ac iona ed hepa in. h ps://doi.o g/10.1371/jou nal.pone.0209110. 001 Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 5 / 15 a enua ion, backsca e and g ayscale in ensi y we e small, wi h na ow limi s o ag eemen be ween obse e s. Fo all es ed pa ame e s, in aobse e a iabili y was sligh ly smalle han in e obse e a ia ion. Limi s o ag eemen o skewness and ku osis o a enua ion, backsca e and g ayscale in ensi y we e wide han o he o he s a is ical pa ame e s. B-A plo o median a enua ion is shown in Fig 2. B-A plo s o mean backsca e and mean g ay- scale in ensi y, espec i ely, a e shown in S3 and S4 Figs. Compa ison o isual TAS and measu emen s by QCU-CMS so wa e The e was an in e se ela ionship be ween se e al QCU-CMS pa ame e s o a enua ion (median alues, = -0.79, p<0.001); 10 h pe cen ile alues = -0.71, p<0.001), backsca e (mean alues = -0.78, p<0.001; 10 h pe cen ile alue = -0.79, p<0.001) and g ayscale in en- si y (mean alues = -0.80, p<0.001; 10 h pe cen ile alues = -0-80, p<0.001)) and he isual TAS. Dis ibu ion o selec ed s a is ical pa ame e s o a enua ion, backsca e and g ayscale in ensi y in he six-s age TAS is shown (Fig 3). Fo he dis ibu ion o he pa ame e s using wo-s age TAS, see S5 Fig. In hese analyses, a minimum h eshold 0.15 mm -1 was used o a enua ion. Table 2. In e obse e a iabili y o QCU-CMS measu emen s. Pa ame e s Obse e 1 Mean (SD) Obse e 2 Mean (SD) Range Mean absolu e di e ence (CI) Mean ela i e di e ence (%) Limi s o ag eemen ICC (CI) A enua ion (mm -1 ) Median 0.61 (0.12) 0.57 (0.13) 0.35–0.77 0.03 (-0.02–0.08) 5.8 -0.11–0.17 0.91 (0.64–0.98) Mean 0.72 (0.09) 0.69 (0.11) 0.49–0.83 0.03 (-0.02–0.08) 4.3 -0.10–0.16 0.88 (0.55–0.97) S anda d de ia ion 0.42 (0.09) 0.42 (0.07) 0.32–0.54 0.001 (-0.02–0.02) 0.2 -0.07–0.07 0.96 (0.84–0.99) Skewness 1.36 (0.50) 1.46 (0.40) 0.89–2.18 -0.10 (-0.26–0.06) -7.2 -0.55–0.19 0.93 (0.73–0.98) Ku osis 3.01 (2.42) 3.72 (2.45) 0.55–7.21 -0.72 (-2.21–0.78) -21.4 -4.90–3.47 0.77 (0.14–0.94) 5 h pe cen ile 0.26 (0.06) 0.23 (0.06) 0.17–0.38 0.02 (-0.01–0.04) 8.6 -0.04–0.08 0.90 (0.52–0.98) 10 h pe cen ile 0.30 (0.07) 0.28 (0.07) 0.19–0.46 0.02 (-0.01–0.05) 8.3 -0.06–0.11 0.90 (0.58–0.97) 90 h pe cen ile 1.27 (0.16) 1.24 (0.17) 1.00–1.53 0.03 (-0.03–0.09) 2.3 -0.14–0.20 0.92 (0.71–0.98) 95 h pe cen ile 1.51 (0.20) 1.48 (0.19) 1.21–1.77 0.03 (-0.04–0.10) 1.9 -0.17–0.23 0.93 (0.75–0.98) Backsca e Median 5.18 (0.22) 5.09 (0.28) 4.77–5.49 0.10 (-0.01–0.20) 1.9 -0.20–0.40 0.87 (0.46–0.97) Mean 5.19 (0.18) 5.10 (0.23) 4.79–5.46 0.10 (0.01–0.18) 1.8 -0.15–0.34 0.86 (0.36–0.97) S anda d de ia ion 0.61 (0.09) 0.66 (0.07) 0.47–0.72 -0.04 (-0.07–-0.02) -6.9 -0.10–0.01 0.89 (-0.09–0.98) Skewness -0.08 (0.29) -0.22 (0.44) -0.74–0.31 0.15 (-0.05–0.34) 96.7 -0.39–0.68 0.83 (0.37–0.96) Ku osis -0.13 (0.49) 0.44 (1.24) -0.70–2.37 -0.57 (-1.16–0.02) -183 -2.22–1.08 0.70 (-0.05–0.92) 5 h pe cen ile 4.22 (0.26) 4.07 (0.26) 3.77–4.70 0.15 (0.05–0.24) 3.5 -0.11–0.41 0.87 (0.05–0.97) 10 h pe cen ile 4.41 (0.25) 4.28 (0.28) 3.95–4.93 0.13 (0.04–0.21) 2.9 -0.12–0.37 0.90 (0.26–0.98) 90 h pe cen ile 5.99 (0.16) 5.94 (0.18) 5.65–6.26 0.05 (-0.02–0.11) 0.8 -0.14–0.23 0.91 (0.65–0.98) 95 h pe cen ile 6.17 (0.16) 6.13 (0.17) 5.85–6.41 0.04 (-0.02–0.10) 0.7 -0.14–0.21 0.92 (0.69–0.98) G ayscale Median 67.19 (32.99) 63.86 (35.43) 28.16–147.83 3.32 (-2.53–9.19) 5.1 -13.05–19.71 0.99 (0.94–1.00) Mean 104.65 (26.87) 100.04 (31.37) 69.76–166.69 4.61 (-2.28–11.49) 4.5 -14.64–23.85 0.97 (0.88–0.99) S anda d de ia ion 106.03 (17.46) 103.02 (15.95) 80.03–134.78 3.02 (-3.78–9.81) 2.9 -15.98–22.01 0.91 (0.67–0.98) Skewness 2.51 (0.88) 2.65 (0.72) 1.72–3.99 -0.14 (-0.41–0.14) -5.3 -0.91–0.63 0.94 (0.76–0.98) Ku osis 12.49 (11.67) 14.62 (10.83) 4.31–38.17 -2.13 (-6.88–2.62) -15.7 -15.40–11.14 0.91 (0.64–0.98) 5 h pe cen ile 16.59 (13.34) 14.74 (11.59) 6.66–48.92 1.85 (0.08–3.61) 11.8 -3.10–6.79 0.99 (0.91–1.00) 10 h pe cen ile 21.77 (16.88) 20.14 (16.17) 8.83–65.09 1.63 (-0.02–3.27) 7.8 -2.97–6.22 0.99 (0.96–1.00) 90 h pe cen ile 239.64 (40.18) 230.74 (48.77) 166.58–312.48 8.90 (-4.40–22.20) 3.8 -28.28–46.09 0.94 (0.80–0.99) 95 h pe cen ile 315.19 (44.98) 305.05 (53.64) 228.63–408.65 10.15 (-6.64–26.94) 3.3 -36.78–57.08 0.94 (0.75–0.98) N = 10. CI, con idence in e al; SD, s anda d de ia ion; ICC, in a-class co ela ion coe icien . h ps://doi.o g/10.1371/jou nal.pone.0209110. 002 Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 6 / 15 The same h ombus c oss-sec ional a eas we e analyzed using a highe minimum h eshold (0.5 mm -1 ) o a enua ion. This esul ed in exclusion o an a e age o 36% (minimum 5.3%, maximum 69.8%) o he pixels o each pa ien p e iously included wi h he 0.15 mm -1 h esh- old. In his analysis, only low pe cen iles o a enua ion alues ( o 10 h pe cen ile alues, = 0.59, p<0.001) had an in e se ela ionship wi h isual TAS. Median ( = -0.05, p = 0.78) o 95 h pe cen ile alues ( = 0.34, p = 0.07) o a enua ion alues we e no s a is ically signi i- can ly ela ed o TAS (S6 Fig). In addi ion, using a io o a enua ion and in ensi y measu emen s, a new ‘no malized’ a - iable was cons uc ed. The a io o 95 h pe cen ile o a enua ion and median in ensi y had a di ec ela ionship wi h TAS ( = 0.86, p<0.001) (S7 Fig). ROC analysis A ROC cu e analysis was applied o assess he bes pa ame e s o dis inguish be ween di e - en ypes o h ombus. The AUCs ep esen ed he abili y o QCU-CMS so wa e pa ame e s o di e en ia e ed and whi e h ombi (Table 4). The bes cu o alues o diagnose ed h ombus Table 3. In aobse e a iabili y o QCU-CMS measu emen s. Pa ame e s Obse e 1 i s ime Mean (SD) Obse e 1 second ime Mean (SD) Range Mean absolu e di e ence (CI) Mean ela i e di e ence (%) Limi s o ag eemen ICC (CI) A enua ion (mm -1 ) Median 0.61 (0.12) 0.59 (0.13) 0.37–0.75 0.02 (-0.03–0.06) 3.3 -0.11–0.14 0.93 (0.74–0.98) Mean 0.72 (0.09) 0.70 (0.11) 0.51–0.84 0.02 (-0.02–0.06) 2.8 -0.10–0.13 0.92 (0.69–0.98) S anda d de ia ion 0.42 (0.09) 0.41 (0.08) 0.31–0.54 0.001 (-0.01–0.02) 0.2 -0.04–0.05 0.98 (0.94–1.00) Skewness 1.36 (0.50) 1.41 (0.43) 0.75–2.14 -0.05 (-0.24–0.14) -3.6 -0.58–0.48 0.92 (0.68–0.98) Ku osis 3.01 (2.42) 3.64 (2.81) 0.52–6.99 -0.64 (-2.48–1.21) -19.2 -5.79–4.52 0.69 (-0.24–0.92) 5 h pe cen ile 0.26 (0.06) 0.25 (0.06) 0.18–0.39 0.01 (-0.01–0.02) 3.9 -0.03–0.05 0.97 (0.90–0.99) 10 h pe cen ile 0.30 (0.07) 0.29 (0.07) 0.20–0.45 0.01 (-0.01–0.03) 3.4 -0.05–0.06 0.97 (0.88–0.99) 90 h pe cen ile 1.27 (0.16) 1.25 (0.17) 1.02–1.54 0.02 (-0.03–0.07) 1.6 -0.12–0.16 0.95 (0.82–0.99) 95 h pe cen ile 1.51 (0.20) 1.48 (0.21) 1.23–1.79 0.03 (-0.03–0.08) 2.0 -0.13–0.18 0.96 (0.86–0.99) Backsca e Median 5.18 (0.22) 5.13 (0.29) 4.78–5.47 0.05 (-0.06–0.16) 1.0 -0.25–0.35 0.91 (0.65–0.98) Mean 5.19 (0.18) 5.14 (0.25) 4.80–5.46 0.04 (0.05–0.13) 0.8 -0.20–0.29 0.92 (0.69–0.98) S anda d de ia ion 0.61 (0.09) 0.62 (0.09) 0.43–0.69 -0.01 (-0.03–-0.01) -1.6 -0.06–0.04 0.98 (0.92–1.00) Skewness -0.08 (0.29) -0.10 (0.31) -0.47–0.29 0.02 (-0.10–0.14) -22.2 -0.32–0.36 0.92 (0.68–0.98) Ku osis -0.13 (0.49) 0.02 (0.56) -0.76–0.86 -0.15 (-0.44–0.13) -136 -0.94–0.64 0.83 (0.37–0.96) 5 h pe cen ile 4.22 (0.26) 4.16 (0.32) 3.78–4.77 0.05 (-0.04–0.14) 1.2 -0.19–0.29 0.95 (0.82–0.99) 10 h pe cen ile 4.41 (0.25) 4.36 (0.31) 3.96–4.94 0.05 (-0.05–0.14) 1.1 -0.21–0.30 0.95 (0.80–0.99) 90 h pe cen ile 5.99 (0.16) 5.96 (0.19) 5.66–6.27 0.03 (-0.03–0.09) 0.5 -0.13–0.19 0.94 (0.79–0.99) 95 h pe cen ile 6.17 (0.16) 6.14 (0.18) 5.86–6.43 0.03 (-0.02–0.08) 0.5 -0.12–0.32 0.95 (0.81–0.99) G ayscale Median 67.19 (32.99) 70.05 (36.06) 30.67–151.52 -2.86 (-8.00–2.28) -4.2 -17.23–11.51 0.99 (0.96–1.00) Mean 104.65 (26.87) 107.50 (29.72) 73.22–170.61 -2.85 (-9.16–3.45) -1.3 -20.48–14.77 0.98 (0.91–0.99) S anda d de ia ion 106.03 (17.46) 107.19 (13.59) 91.86–136.54 -1.16 (-7.88–5.56) -1.1 -19.95–17.64 0.91 (0.63–0.98) Skewness 2.51 (0.88) 2.52 (0.70) 1.63–3.99 -0.003 (-0.33–0.32) -0.1 -0.91–0.90 0.92 (0.70–0.98) Ku osis 12.49 (11.67) 12.84 (10.82) 3.81–38.65 -0.35 (-5.00–4.30) -2.8 -13.35–12.65 0.92 (0.66–0.98) 5 h pe cen ile 16.59 (13.34) 17.19 (14.43) 7.17–54.29 -0.60 (-1.97–0.76) -3.6 -4.41–3.21 1.00 (0.98–1.00) 10 h pe cen ile 21.77 (16.88) 22.96 (18.30) 9.50–69.53 -1.20 (-2.97–0.58) -5.4 -6.15–3.76 0.99 (0.98–1.00) 90 h pe cen ile 239.64 (40.18) 244.76 (38.86) 195.65–319.26 -5.12 (-19.99–9.75) -2.1 -46.69–36.46 0.93 (0.73–0.98) 95 h pe cen ile 315.19 (44.98) 320.95 (41.95) 273.50–414.96 -5.76 (-24.41–12.90) -1.8 -57.91–46.39 0.91 (0.63–0.98) N = 10. CI, con idence in e al; SD, s anda d de ia ion; ICC, in a-class co ela ion coe icien . h ps://doi.o g/10.1371/jou nal.pone.0209110. 003 Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 7 / 15 in each class o a iables we e <5.35 o mean backsca e (AUC = 0.959, sensi i i y 92%, spec- i ici y 94%), <120.1 o mean g ayscale in ensi y (AUC = 0.959, sensi i i y 85%, speci ici y 100%), <0.57 mm -1 o median a enua ion (AUC = 0.919, sensi i i y 100%, speci ici y 71%) and >0.022 mm -1 o 95 h pe cen ile o a enua ion/median g ayscale in ensi y (AUC = 0.950, sensi i i y 88%, speci ici y 85%) (Fig 4). In an addi ional analysis, using bina y logis ic eg es- sion, bes p edic o s o h ombus ype we e unchanged. AUCs o di e en s a is ical a iables o a enua ion measu emen s om he second analy- sis using a highe 0.5 mm -1 h eshold we e low (in ange o 0.50–0.60) excep o 5 h pe cen ile and 10 h pe cen ile o a enua ion (AUCs 0.878 and 0.851, espec i ely). Fig 2. In a- and in e obse e a iabili y o measu emen o h ombus a enua ion in OCT images using image analysis so wa e. Sca e plo (le ) and Bland- Al man plo ( igh ) o in aobse e (A) and in e obse e (B) compa ison o median a enua ion. OCT, op ical cohe ence omog aphy; SD, s anda d de ia ion. h ps://doi.o g/10.1371/jou nal.pone.0209110.g002 Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 8 / 15 Discussion As a i s s ep in o de o esol e he clinical signi icance and p edic i e po en ial o OCT- based h ombus analysis in clinical pa ien s, a obus me hod o ansla e he adi ional isual classi ica ion in o ep oducibly measu ed a iables is needed. Valida ed me hodology o assess di e en h ombus ypes by OCT imaging has no been a ailable, and p e iously published da a has elied on subjec i e analysis me hods based on obse e s’ isual pe cep ion o he h ombus ype [9,10]. In his me hodological s udy, we measu ed a enua ion, backsca e and g ayscale in ensi y o in i o h ombi using newly de eloped image analysis so wa e ea u es. Fo me hodological alida ion, we assessed epea abili y o he measu emen s and he co ela- ion be ween adi ional isual h ombus classi ica ion and he compu e so wa e-based anal- ysis. We we e able o demons a e o he i s ime he use o image analysis so wa e o assess in i o h ombi in OCT images. Measu emen s o a enua ion, backsca e and g ayscale in en- si y o he h ombi using he so wa e we e highly epea able and he e was good ag eemen be ween wo obse e consensus isual e alua ion o h ombus ype and se e al pa ame e s measu ed using he so wa e on he OCT images. Visual assessmen o h ombus ype in OCT images is pe se highly obse e -dependen . The e is no much p e ious da a on he epea abili y o he ope a o -based assessmen o h ombus ype and he e is also a lack o s anda dized de ini ions and cu -o alues di e en i- a ing ed and whi e h ombus. We obse ed ha he isual classi ica ion, as expec ed, was only mode a ely ep oducible. O no e, he p esen analysis was pe o med o -line by expe ienced analys s and ep oducibili y ou side he OCT co e lab se ing is mos p obably e en lowe . In he p esen s udy, we showed ha compu e -based image analysis o in i o h ombus can be Fig 3. Rela ionship o h ombus a enua ion sco e and pa ame e s measu ed by image analysis so wa e in h ombus a eas in OCT images. Sca e plo s o median a enua ion (A), 10 h pe cen ile o a enua ion (B), mean backsca e (C), 10 h pe cen ile o backsca e (D), mean g ayscale in ensi y (E) and 10 h pe cen ile o g ayscale in ensi y (F). OCT, op ical cohe ence omog aphy. h ps://doi.o g/10.1371/jou nal.pone.0209110.g003 Analysis o in aco ona y h ombus by OCT PLOS ONE | h ps://doi.o g/10.1371/jou nal.pone.0209110 Decembe 17, 2018 9 / 15