Follicular thyroid lesions: Is there a discriminatory potential in the computerized nuclear analysis?

7:8 907–913F O Valen im e al. Thy oid ollicula lesions diagnosis RESEARCH Follicula hy oid lesions: is he e a disc imina o y po en ial in he compu e ized nuclea analysis? Flá iaO Valen im1, Bá ba aP Coelho1, HélioA Mio 2, Ca olineY Hayashi1, DaniloTA Jaune1, C is ianoC Oli ei a3, Ma iângelaEA Ma ques3, JoséVicen e Taglia ini4, EmanuelC Cas ilho4, Paula Soa es5,6,7 and GláuciaMFS Maze o1 1In e nal Medicine Depa men , Bo uca u Medical School, Sao Paulo S a e Uni e si y (Unesp), Bo uca u, São Paulo, B azil 2Depa men o De ma ology, Bo uca u Medical School, Sao Paulo S a e Uni e si y (Unesp), Bo uca u, São Paulo, B azil 3Pa hology Depa men , Bo uca u Medical School, Sao Paulo S a e Uni e si y (Unesp), Bo uca u, São Paulo, B azil 4O ola yngology and Head and Neck Su ge y Depa men , Bo uca u Medical School, Sao Paulo S a e Uni e si y (Unesp), Bo uca u, São Paulo, B azil 5i3S - Ins i u o de In es igação e Ino ação em Saúde, Uni e sidade do Po o, Po o, Po ugal 6Cance Signaling and Me abolism G oup, Ins i u e o Molecula Pa hology and Immunology o he Uni e si y o Po o (IPATIMUP), Po o, Po ugal 7Depa men o Pa hology, Medical Facul y, Uni e si y o Po o, Po o, Po ugal Co espondence should be add essed o G M F S Maze o: [email p o ec ed] Abs ac Backg ound: Compu e ized image analysis seems o ep esen a p omising diagnos ic possibili y o hy oid umo s. Ou aim was o e alua e he disc imina o y diagnos ic e iciency o compu e ized image analysis o cell nuclei om his ological ma e ials o ollicula umo s. Me hods: We s udied pa a in-embedded ma e ials om 42 ollicula adenomas (FA), 47 ollicula a ian s o papilla y ca cinomas (FVPC) and 20 ollicula ca cinomas (FC) by he so wa e ImageJ. Based on he nuclea mo phome y and ch oma in ex u e, he samples we e classi ied as FA, FC o FVPC using he Classi ica ion and Reg ession T ees me hod. Resul s: We obse ed high diagnos ic sensi i i y and speci ici y a es (FVPC: 89.4% and 100%; FC: 95.0% and 92.1%; FA: 90.5 and 95.5%, espec i ely). When he umo s we e compa ed by pai s (FC s FA, FVPC s FA), 100% o he cases we e classi ied co ec ly. Conclusion: The compu e ized image analysis o nuclea ea u es showed o be a use ul diagnos ic suppo ool o he his ological di e en ia ion be ween ollicula adenomas, ollicula a ian s o papilla y ca cinomas and ollicula ca cinomas. In oduc ion The di e en ial diagnosis o hy oid ollicula lesions includes bo h benign umo s, such as ollicula adenoma (FA) and malignan umo s, such as ollicula ca cinoma (FC) and ollicula a ian o papilla y ca cinoma (FVPC). This diagnosis can be di icul , especially conside ing FA and FC, in which he dis inc ion depends on de ec ing ascula and/o capsula in asion and demands ho ough e alua ion o he his ological ma e ial (1). Thus, bo h incomple e lesion excision, lacking a sample o he in il a ed umo capsule po ion, and he ex a ime equi ed o assess he en i e umo ex ension hinde he di e en ial diagnosis be ween FA and FC (2) and can make i un easible. Changes in nuclea mo phome y and ch oma in ex u e a e classically desc ibed by pa hologis s as ac o s ha di e en ia e no mal issues om neoplasms. -18-0237 Key Wo ds adenoca cinoma ollicula ca cinoma papilla y cell nucleus his ology hy oid neoplasms Endoc ine Connec ions (2018) 7, 907–913 ID: 18-0237 78 This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License. h ps://doi.o g/10.1530/EC-18-0237 h p://www.endoc ineconnec ions.o g © 2018 The au ho s Published by Bioscien i ica L d Downloaded om Bioscien i ica.com a 01/13/2020 05:40:14PM ia ee access F O Valen im e al. Thy oid ollicula lesions diagnosis 9087:8 Mo phological analysis o cell nuclei by his ology may p o ide da a abou cellula physiology and con ibu e o he es ablishmen o he diagnosis and p ognosis o neoplas ic lesions (3). Compu e ized image analysis is an objec i e and highly ep oducible ool in daily p ac ice and has p o en o be a p omising esou ce o assessing hy oid lesions. Mo e widely used o assessing his ological ma e ial (1, 4, 5, 6, 7, 8, 9, 10), compu e ized image analysis enables he de ec ion o e y sub le nuclea changes (11), whose in es iga ion could also p o ide da a o cy ological ma e ial analysis (11, 12, 13). Howe e , speci ically o ollicula -pa e ned umo s, nuclea pa ame e s ha can co ec ly di e en ia e malignan om benign lesions ha e no ye been es ablished. This could be pa ly due o he small numbe o samples analyzed in p e ious s udies, o absence o no maliza ion o he esul s by nuclea dimension and o he assessmen o ew nuclea pa ame e s (1, 11, 13). Hence, he objec i e o his s udy was o e alua e he disc imina o y diagnos ic e iciency o compu e ized image analysis o cell nuclei in his ological ma e ials ob ained om FA, FC and FVPC. Ma e ials and me hods Design, pa ien s and his ological ma e ial p ocessing This c oss-sec ional s udy used compu e ized image analysis o compa e he nuclea mo phome y and ch oma in ex u e ea u es o hy oid umo issues and heal hy hy oid issue (Fig.1) and o assess i s diagnos ic accu acy. When making models using machine lea ning me hods, o e i ing should always be conside ed. The e o e, he ideal in hose cases would be o e alua e a la ge sample and o di ide i in o a aining g oup, which de elops he e alua ion algo i hm, and a alida ion g oup, which applies he de eloped algo i hm and measu es i s pe o mance. This is a p elimina y s udy, in which he aining was ca ied ou esul ing in he de elopmen o he algo i hm o be alida ed la e . S o ed his ological ma e ials o 103 pa ien s wi h his opa hological diagnoses o FA, FC and FVPC we e analyzed. The pa ien s had unde gone o al hy oidec omy be ween 1993 and 2016 a he Clinics Hospi al o Bo uca u School o Medicine. The s udy popula ion was i s cha ac e ized by gende and age. Nex , wo expe ienced pa hologis s (MEAM and CCO) assessed he hy oid issue o con i m he diagnoses based on his opa hological pa ame e s (14). A o al o 109 umo s om 103 pa ien s we e included: 42 FA, 47 FVPC and 20 FC (six pa ien s p esen ed wo lesions: i e had FA and FVPC and one had FA and FC). Tissue adjacen o he benign umo s (FA) was used as ep esen ing no mal hy oid issue (NT). Th ee mic on sec ions we e cu om pa a in-embedded ep esen a i e issue samples and s ained wi h hema oxylin-eosin (HE). This s udy was app o ed by he Resea ch E hics Commi ee o Bo uca u School o Medicine – Unesp – unde p o ocol numbe 635,459, issued on May 05, 2014. Image acquisi ion The slides we e scanned and pho og aphed (43× magni ica ion; PANORAMIC MIDI II – 3D His ech, Japan; h p://3dhis ech.com/panno amic_midi) and numbe ed andomly. Two o h ee pho os we e cap u ed Figu e1 Thy oid issues analyzed: no mal hy oid (A), ollicula adenoma (B), ollicula a ian o papilla y ca cinoma (C), and ollicula ca cinoma (D). Hema oxylin-eosin s aining, ×20. This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License. h ps://doi.o g/10.1530/EC-18-0237 h p://www.endoc ineconnec ions.o g © 2018 The au ho s Published by Bioscien i ica L d Downloaded om Bioscien i ica.com a 01/13/2020 05:40:14PM ia ee access F O Valen im e al. Thy oid ollicula lesions diagnosis 9097:8 om each slide. Thy oid neoplasms may be qui e he e ogeneous and choosing a slide a ea o pho og aph could esul in di e en nuclea ep esen a i eness. The e o e, he pho og aphs we e cap u ed om he mos ep esen a i e a eas p esen ing a highe numbe o nuclei, and he bes pho o o each slide was chosen o be analyzed, based on he judgmen o wo expe ienced pa hologis s (MEAM and CCO). Image analysis The images we e analyzed by wo esea che s who did no know he his opa hological diagnosis o he samples (blind analysis). The diame e s o he umo s in each slide we e measu ed in cen ime e s, and he nuclei we e analyzed by he ee so wa e ImageJ (h p://imagej.ne / Downloads). The so wa e u ns colo images in o black and whi e images con aining 256 shades o g ay (16 bi s), which allows be e analysis o ch oma in compac ion and elimina es colo -based di e ences be ween he samples (10). The nuclei in he edi ed pho og aphs we e manually selec ed and analyzed indi idually. Agglome a ed o o e lapping nuclei we e excluded. A pilo s udy de e mined he minimum numbe o nuclei ha should be analyzed in each pho og aph – any hing abo e ha minimum would no a ec he esul s o he s a is ical analysis. Six y nuclei pe pho og aph would be su icien (da a no shown) based on a s anda d e o o 5% o each a iable. Thus, a o al o 9060 nuclei we e assessed, including he NT samples. Fo each selec ed nucleus, he p og am p o ides a lis o mo phome ic and ex u e pa ame e s, allowing he use o selec he pa ame e s o in e es o nuclea assessmen (16). We also calcula ed seconda y indica o s ( a ia ion coe icien s) ela ed o hese pa ame e s. Hence, mo phome ic assessmen included: a ea, pe ime e s, ci cula i y, la ge diame e (Fe e ), a io be ween la ge and smalle diame e (Aspec Ra io, AR), index a ea pe Fe e , a io be ween pe ime e and a ea (P/A), a io be ween a ea and Fe e (a ea/Fe e ) and index pe ime e / a ea. The selec ed ex u e ea u es we e mean and median g ay in ensi y (exp essed in a scale o 256 shades o g ay, in which highe numbe s mean ligh e nuclei: 0 = black; 255 = whi e), s anda d de ia ion o g ay in ensi y (STDEV), oughness (RA), egula i y o nuclea memb ane (Round), solidi y, ac al dimension (F ac al), en opy and a io be ween g ay in ensi y and a ea (g ay in ensi y/a ea). The calcula ed seconda y indica o s we e coe icien o a ia ion (CV) o a ea, CV-Mean in ensi y, CV-STDEV, CV-Pe ime e , CV-Ci cula i y, CV-Fe e , CV-Median in ensi y, CV-AR, CV-Round, CV-Solidi y, CV-F ac al, CV-En opy, CV-Pe ime e /a ea and CV-RA. The so wa e exp essed hese pa ame e s as pixels, which we e la e ans o med in o mic ons. Compa ison and classi ica ion o he umo s The umo and he heal hy hy oid issues we e ini ially compa ed o mo phome ic and ex u al aspec s o ch oma in. La e , umo s wi h his ological diagnoses o FA, FC and FVPC we e analyzed o hese nuclea pa ame e s wi h he Classi ica ion and Reg ession T ees (CRT) me hod, which p o ided a eclassi ica ion o hem. The diagnosis p o ided by he CRT was hen compa ed o he his ological diagnosis, hus e alua ing he me hod's abili y o co ec ly classi y hese lesions. Since he majo doub when acing a ollicula pa e n umo esides in di e en ia ing benign om malignan lesions, he umo s we e also analyzed by pai s (malign s benign: FC s FA, FVPC s FA). S a is ical analysis The equencies o emales by umo diagnosis we e compa ed by he chi-squa e es . The pa ien s’ ages, g ea e umo diame e and nuclea pa ame e s we e exp essed as medians and qua iles (p25–p75) and compa ed by he K uskal–Wallis es . Based on he s udy nuclea ea u es, he umo s we e eclassi ied by he CRT me hod and woing algo i hm (impu i y measu e). This s a is ical me hod cons uc s a ee model, wi h p edic i e cha ac e o he da a, ha s a s om a oo and b anches ou , om nodes ha use cu -o s o he pa ame e s analyzed, p o ided by he model i sel (17). The sensi i i y, speci ici y and a ea unde he ecei e ope a o cha ac e is ic (ROC) cu e (AUC) we e calcula ed o each umo o de e mine he alidi y o he classi ica ion me hod. The a ious nuclea mo phome ic and ex u e pa ame e s we e also co ela ed wi h g ea e umo diame e using Spea man’s ho coe icien . The so wa e IBM SPSS/Windows ( e sion 22) pe o med all s a is ical analyses using a signi icance le el o 5%. Resul s FC pa ien s we e olde (P = 0.04) han FVPC pa ien s and hei umo s had la ge diame e s (P = 0.001) han FVPC. The FA did no di e om he o he umo s in hese pa ame e s. No signi ican di e ences we e p esen This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License. h ps://doi.o g/10.1530/EC-18-0237 h p://www.endoc ineconnec ions.o g © 2018 The au ho s Published by Bioscien i ica L d Downloaded om Bioscien i ica.com a 01/13/2020 05:40:14PM ia ee access F O Valen im e al. Thy oid ollicula lesions diagnosis 9107:8 conce ning he gende dis ibu ion acco ding o diagnosis (P = 0.41) (Table 1). Tumo diame e was no associa ed wi h he s udy mo phome ic and ex u e pa ame e s (Supplemen a y Table 1, see sec ion on supplemen a y da a gi en a he end o his a icle), excep o he CV-STDEV pa ame e s, which we e weakly nega i ely co ela ed ( = −0.24; P = 0.02). The e o e, he esul s we e no co ec ed by umo diame e . All he s udied p ima y nuclea pa ame e s di e ed signi ican ly (P < 0.05) (Table 2 and Supplemen a y Table 2) be ween he di e en lesions, bu no he seconda y indica o s ( a ia ion coe icien s) CV-A ea (P = 0.08), CV-STDEV (P = 0.24), CV-Fe e (P = 0.22), and CV-AR (P = 0.05) (Supplemen a y Table2). When he nuclea pa ame e s o he h ee umo s we e analyzed by he CRT eg ession model (Fig.2) based on he CV-RA pa ame e s, he esul ing global classi ica ion ma ched he ac ual diagnoses in 90.8% o he cases, wi h sensi i i ies o 95, 90.5 and 89.4% o FC, FA and FVPC, espec i ely (Supplemen a y Table 3). The speci ici ies and AUC we e 92.1–100% and 0.93–0.95, espec i ely (Table3). When only FC and FA we e analyzed, co ec global classi ica ion was achie ed o 100% o he umo s (Fig.3 and Supplemen a y Table4). Likewise, when only FVPC and FA we e analyzed, co ec global classi ica ion was also achie ed o 100% o he cases (Supplemen a y Fig.1 and Supplemen a y Table5). Discussion Nume ous s udies ha e a emp ed o de e mine an e icien and accu a e diagnos ic me hod o ollicula - pa e ned hy oid nodules (2). One o he echniques ha has been used o his pu pose is compu e ized image analysis o cell nuclei (4, 5, 7, 8, 12, 18, 19, 20, 21), which has p o en p omising bu no en i ely conclusi e. This p elimina y s udy uses compu e ized image analysis o cell nuclei and a classi ica ion me hod, which p o ed capable o co ec ly iden i ying high pe cen ages o FA, FC and FVPC by image analysis ob ained om he his ological ma e ials. Assessmen o hese h ee ollicula -pa e ned umo s enabled us o co ec ly globally classi y mo e han 90% o he cases. Addi ionally, high sensi i i ies (be ween 89.4% and 95%), Table 1 Gene al da a o 109 umo s in 103 pa ien s. Pa ame e Diagnosis o he umo s P FA FVPC FC Pa ien s* (n) 36 47 20 – Women (n (%)) 34 (94.4) 44 (93.6) 15 (75.0) 0.41 Age (yea s)†54 (44; 60)ab 52 (39; 59)ª 61 (46; 73)b0.04 La ge diame e (cm)† 2.5 (1.2; 3.5)ab 1.5 (1; 3)a 3.5 (3; 4.2)b 0.001 *Only he wo s diagnosis was conside ed. †Median (25 h pe cen ile, 75 h pe cen ile). S a is ical es : K uskal–Wallis. Di e en le e s mean s a is ical di e ence (b > a; P < 0.05). FA, ollicula adenoma; FC, ollicula ca cinoma; FVPC, ollicula a ian o papilla y ca cinoma. Table 2 Mo phome ic and ex u al p ima y pa ame e s, acco ding o he his ological ma e ial e alua ed. Pa ame e His opa hologic ma e ial* P** NT FA FVPC FC A ea (µm2) 40 (34.77; 46.2) 44.91 (38.15; 55.2) 59.73 (46.91; 73.16) 64.35 (50.56; 69.54) 0.00 Mean g ay in ensi y†59.1 (53.05; 65.26) 55.35 (45.53; 68.06) 83.71 (66.31; 94.92) 78.04 (57.91; 85.31) 0.00 STDEV 41.15 (36.6; 45.1) 34.6 (29.31; 41.85) 44.28 (36.45; 51.54) 38.17 (33.34; 43.54) 0.00 Pe ime e (µm) 24.38 (23.15; 26.82) 26.59 (24.17; 29.69) 30.22 (27.62; 34.73) 32.55 (28.01; 35.13) 0.00 Ci cula i y 0.85 (0.8; 0.87) 0.83 (0.8; 0.86) 0.79 (0.7; 0.87) 0.78 (0.72; 0.84) 0.00 Fe e (µm) 8.74 (8.35; 9.85) 9.31 (8.73; 10.39) 10.66 (9.88; 12.15) 11.14 (9.47; 12.01) 0.00 Median g ay in ensi y†47 (39; 53.5) 43.25 (38; 58.5) 74.5 (57.5; 86.5) 71.25 (48; 81.25) 0.00 AR 1.38 (1.27; 1.46) 1.27 (1.24; 1.39) 1.32 (1.26; 1.41) 1.28 (1.23; 1.35) 0.02 Round 0.73 (0.68; 0.78) 0.78 (0.72; 0.81) 0.75 (0.7; 0.79) 0.78 (0.74; 0.81) 0.02 Solidi y 0.93 (0.92; 0.94) 0.93 (0.92; 0.94) 0.92 (0.89; 0.94) 0.92 (0.9; 0.93) 0.04 F ac al 2.36 (2.33; 2.38) 2.39 (2.36; 2.43) 2.46 (2.43; 2.48) 2.43 (2.34; 2.44) 0.00 En opy 5.27 (4.76; 5.73) 4.68 (2.27; 5.17) 4.65 (4.34; 5.04) 4.35 (4.06; 4.61) 0.00 Pe ime e /a ea (µm) 0.6 (0.58; 0.66) 0.58 (0.53; 0.63) 0.5 (0.47; 0.62) 0.52 (0.49; 0.56) 0.00 RA 130.9 (124.4; 142.2) 133.8 (121.6; 152.2) 105.0 (93.1; 123.4) 107.8 (98.7; 135.8) 0.00 A ea/ e e (µm) 4.52 (4.11; 4.79) 4.93 (4.36; 5.21) 5.62 (4.58; 6.01) 5.58 (5.22; 6.05) 0.00 G ay in ensi y/a ea (unidades/µm2) 0.13 (0.12; 0.16) 0.12 (0.1; 0.14) 0.12 (0.12; 0.15) 0.12 (0.1; 0.14) 0.02 *Median (pe cen ile 25; pe cen ile 75). **Signi ican di e ences (be ween he ou g oups): P < 0.05; s a is ical es : K uskal–Wallis. †Exp essed in a scale o 256 shades o g ay, in which highe numbe s mean ligh e nuclei (0 = black; 255 = whi e). AR, Aspec Ra io; FA, ollicula adenoma; FC, ollicula ca cinoma; FVPC, ollicula a ian o papilla y ca cinoma; µm: mic ome e ; NT, no mal hy oid; RA, oughness; STDEV, s anda d de ia ion o g ay in ensi y. This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License. h ps://doi.o g/10.1530/EC-18-0237 h p://www.endoc ineconnec ions.o g © 2018 The au ho s Published by Bioscien i ica L d Downloaded om Bioscien i ica.com a 01/13/2020 05:40:14PM ia ee access F O Valen im e al. Thy oid ollicula lesions diagnosis 9117:8 speci ici ies (be ween 92.1% and 100%) and AUC (be ween 0.93 and 0.95) we e ob ained when he assessmen me hod was used o classi ying umo ype. Gi en he diagnos ic unce ain y be ween FC and FA (2), we analyzed he alidi y o he classi ica ion me hod o hese wo umo s and co ec ly diagnosed 100% o he samples. Simila ly, when only FVPC and FA we e assessed, 100% o he samples we e classi ied co ec ly. O he s udies ha used compu e ized nuclea image analysis o assess hy oid umo s epo ed co ec global classi ica ion a es o 62–100% (1, 5). The easons o classi ica ion a e a iabili y may in ol e he sample size, he numbe and ype o he nuclea pa ame e s e alua ed and he in e e ence o o he ac o s, such as hose ela ed o he umo s s udied. Indeed, small sample sizes and assessmen o ew nuclea pa ame e s may a ec he esul s (1). The p esen s udy assessed 109 umo s, a easonably obus numbe , and 33 nuclea pa ame e s, including mo phome ic and ex u e ea u es, and seconda y nuclea indica o s, which enabled mo e accu a e analysis o each nucleus, and consequen ly, o each umo . Appa en ly, he a e o co ec umo diagnosis inc eases wi h he numbe o s udied pa ame e s (7, 11). Some nuclea pa ame e s included in he p esen s udy ha e al eady been used by o he au ho s and p o ed p omising o he inal cha ac e iza ion o he a ious ollicula lesions (4, 5, 8, 19, 22). None heless, s udies using seconda y indica o s (CV) we e no ound. In he p esen s udy, seconda y indica o s con ibu ed o he high pe cen ages o co ec classi ica ion in he CRT eg ession model. O he p oblems ha could a ec he esul s a e he inclusion o high umo sub ype di e si y and dis ega ding he ac o umo size, ha is, no co ec ing nuclea pa ame e s by size (1, 7, 11), which could limi he use ulness o he me hod o lesions o ce ain diame e s (11, 13). The p esen s udy assessed only ollicula -pa e ned umo s. Al hough FCs had g ea e diame e s, umo dimension was no associa ed wi h nuclea pa ame e s, sugges ing ha umo size does no in luence nuclea a iables. Thus, ou esul s sugges ha he me hod may be used in hy oid umo s ega dless o hei dimensions. Figu e2 Classi ica ion by he me hod Classi ica ion and Reg ession T ees (CRT), wi h algo i hm o woing, o he ollicula adenomas (FA), ollicula ca cinomas (FC) and ollicula a ian o papilla y ca cinomas (FVPC), om he e alua ed nuclea pa ame e s. The numbe s and le e s o he low cha a e explained in he able. AR, Aspec Ra io; CV, coe icien o a ia ion; DX, diagnosis; RA, oughness; PERIM, Pe ime e ; STDEV, s anda d de ia ion o g ay in ensi y. Table 3 Validi y o he use o compu ed nuclea mo phome y analysis, wi h he Classi ica ion and Reg ession T ees (CRT) analysis, as a classi ica o y me hod. Tumo s FA FVPC FC Sensi i i y (%) 90.5 89.4 95.0 Speci ici y (%) 95.5 100.0 92.1 PPV (%) 92.7 100.0 73.1 NPV (%) 94.1 92.5 98.8 A ea unde he ROC cu e 0.93 0.95 0.94 FA, ollicula adenomas; FC, ollicula ca cinomas; FVPC, ollicula a ian o papilla y ca cinoma; NPV, nega i e p edic i e alue; PPV, posi i e p edic i e alue. This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License. h ps://doi.o g/10.1530/EC-18-0237 h p://www.endoc ineconnec ions.o g © 2018 The au ho s Published by Bioscien i ica L d Downloaded om Bioscien i ica.com a 01/13/2020 05:40:14PM ia ee access F O Valen im e al. Thy oid ollicula lesions diagnosis 9127:8 This s udy has some limi a ions. A i s limi a ion would be he applicabili y o he echnique, which is somewha ime demanding as i equi es nuclea image analysis. Howe e , he ime equi ed o such ask a ies be ween indi iduals and dec eases wi h expe ience, un il i pla eaus a abou 30 min pe case in ou expe ience. This s udy’s esul s could encou age he de elopmen o au oma ed sys ems o op imize slide selec ion ime and analysis. Ano he limi a ion would be he size o ees and numbe o kno s esul ing om CRT model analysis. La ge ees may no be use ul and may be associa ed wi h o e i ing, he eby p o iding no explana o y powe . Howe e , he ees p esen ed he e could be conside ed only o mode a e size. Mo eo e , as dis inguishing umo s h ough pa hological cha ac e is ics is di icul , i is expec ed ha he a iables o igina ed om hose cha ac e is ics migh also be di icul o e alua e, which in ol es a g ea e numbe o kno s. Ano he limi a ion ega ds he sample size, which could s ill be conside ed no la ge enough, al hough o he s udies ha e al eady ob ained meaning ul esul s wi h e en smalle sample sizes (1). In addi ion, he numbe o nuclei assessed was based on a pilo s udy ha de e mined he minimum sample size, and his pa ame e can be u he op imized. I is impo an o emphasize ha his is a p elimina y s udy, in which he e alua ion algo i hm was de eloped. The nex s ep will consis o selec ing a much la ge sample o alida e he obse ed esul s. Despi e he limi a ions lis ed abo e, he high sensi i i y and speci ici y o he me hod used by he p esen s udy makes i an impo an ool o aiding he his ological diagnosis o ollicula -pa e ned cases. In addi ion, hese p elimina y esul s s imula e he implemen a ion o u u e s udies wi h dis inc designs, which could include mo e umo sub ypes, such as he ecen ly p oposed NIFTP (nonin asi e ollicula hy oid neoplasm wi h papilla y-like nuclea ea u es) (23, 24); he explo a ion o clinical, labo a o ial, and p ognos ic a iables (25); and he e alua ion o nuclea mo phome ic and ch oma in ex u e cha ac e is ics also o he ollicula lesions om cy ological samples. In conclusion, compu e ized image analysis o nuclea ea u es showed o be a p omising diagnos ic suppo ool o he his ological di e en ia ion be ween FA, FC and FVPC, wi h high sensi i i y and speci ici y. The e o e, his inexpensi e, ep oducible and ope a o - independen me hod migh ep esen a aluable diagnos ic suppo ool o he assessmen o ollicula - pa e ned umo s and pa icula ly, o di e en ia ing FA om FC. Howe e , due o he conside a ions and limi a ions men ioned abo e, he p elimina y esul s p esen ed in his s udy should ye be alida ed, and hen con i med by independen g oups. Mo eo e , o he s udies a e needed o assess he applicabili y o hese indings in cy ological ma e ials wi h inde e mina e diagnosis. Figu e3 Classi ica ion by he me hod Classi ica ion and Reg ession T ees (CRT), wi h algo i hm o woing, o he ollicula adenomas (FA) and ollicula ca cinomas (FC), om he e alua ed nuclea pa ame e s. The numbe s and le e s o he low cha a e explained in he able. AR, Aspec Ra io; CIRC, ci cula i y; CV, coe icien o a ia ion; DX, diagnosis; PERIM, pe ime e . This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License. h ps://doi.o g/10.1530/EC-18-0237 h p://www.endoc ineconnec ions.o g © 2018 The au ho s Published by Bioscien i ica L d Downloaded om Bioscien i ica.com a 01/13/2020 05:40:14PM ia ee access F O Valen im e al. Thy oid ollicula lesions diagnosis 9137:8 Supplemen a y da a This is linked o he online e sion o he pape a h ps://doi.o g/10.1530/ EC-18-0237. Decla a ion o in e es The au ho s decla e ha he e is no con lic o in e es ha could be pe cei ed as p ejudicing he impa iali y o he esea ch epo ed. Funding This s udy ecei ed inancial suppo om Fundação de Ampa o à Pesquisa do Es ado de São Paulo (FAPESP; p ocess numbe 2014/10028-2), and PIBIC/PROPE-Unesp (p ocess numbe 33347). Acknowledgmen s The au ho s hank o Fundação de Ampa o à Pesquisa do Es ado de São Paulo (FAPESP; p ocess numbe 2014/10028-2), and PIBIC/PROPE-Unesp (p ocess numbe 33347) o he esea ch suppo and o Ma cos Robe o F anchi and Luiz Fe nando F anchi o he help in p ocessing he his ological ma e ial. Re e ences 1 WangW, OzolekJA & RohdeGK. De ec ion and classi ica ion o hy oid ollicula lesions based on nuclea s uc u e om his opa hology images. Cy ome y 2010 77 485–494. (h ps://doi. o g/10.1002/cy o.a.20853) 2 He essCS & ThompsonLD. Minimally in asi e ollicula hy oid ca cinoma. Endoc ine Pa hology 2001 12 417–422. (h ps://doi. o g/10.1385/EP:12:4:417) 3 MendaçolliPJ, B ianeziG, Schmi JV, Ma quesME & Mio HA. Nuclea mo phome y and ch oma in ex u al cha ac e is ics o basal cell ca cinoma. Anais B asilei os de De ma ologia 2015 90 874–878. (h ps://doi.o g/10.1590/abd1806-4841.20154076) 4 F asolda iA, Flo aM, Pesen iM, Ca oggioA & Valca ilR. Compu e - assis ed cell mo phome y and ploidy analysis in he assessmen o hy oid ollicula neoplasms. Thy oid 2001 11 941–946. (h ps://doi. o g/10.1089/105072501753211000) 5 Gup aN, Sa ka C, SinghR & Ka akAK. E alua ion o diagnos ic e iciency o compu e ized image analysis based quan i a i e nuclea pa ame e s in papilla y and ollicula hy oid umo s using pa a in- embedded issue sec ions. Pa hology and Oncology Resea ch 2001 7 46–55. (h ps://doi.o g/10.1007/BF03032605) 6 Mu a aSI, MochizukiK, NakazawaT, KondoT, Nakamu aN, Yamashi aH, U a aY, Ashiha aT & Ka ohR. Mo phological abs ac ion o hy oid umo cell nuclei using mo phome y wi h ac o analysis. Mic oscopy Resea ch and Technique 2003 61 457–462. (h ps://doi.o g/10.1002/jem .10355) 7 Demi C & Yene B. Au oma ed cance diagnosis based on his opa hological images: a sys ema ic su ey. In Technical Repo cTR-05-09, Rensselae Poly echnic Ins i u e, Depa men o Compu e Science. New Yo k, NY, USA: Rensselae Poly echnic Ins i u e, 2005. 8 Ka sliogluY, CelasunB & GunhanO. Con ibu ion o mo phome y in he di e en ial diagnosis o ine-needle hy oid aspi a es. Cy ome y 2005 65 22–28. (h ps://doi.o g/10.1002/cy o.b.20024) 9 P oie iA, Sa o iC, Bo elliN, GianniniR, Ma e azziG, Leoca aP, EliseiR, Vi iP, MiccoliP & BasoloF. Follicula -de i ed neoplasms: mo phome ic and gene ic di e ences. Jou nal o Endoc inological In es iga ion 2010 36 1055–1061. (h ps://doi.o g/10.3275/9063) 10 JungC & KimC. Impac o he accu acy o au oma ic segmen a ion o cell nuclei clus e s on classi ica ion o hy oid ollicula lesions. Cy ome y 2014 85 709–718. (h ps://doi.o g/10.1002/cy o.a.22467) 11 Mu a aSI, MochizukiK, NakazawaT, KondoT, Nakamu aN, Yamashi aH, U a aY, Ashiha aT & Ka ohR. De ec ion o unde lying cha ac e is ics o nuclea ch oma in pa e ns o hy oid umo cells using ex u e and ac o analyses. Cy ome y 2002 49 91–95. (h ps:// doi.o g/10.1002/cy o.10162) 12 Rodenacke K & Beng ssonE. A ea u e se o cy ome y on digi ized mic oscopy images. Analy ical Cellula Pa hology 2003 25 1–36. (h ps://doi.o g/10.1155/2003/548678) 13 Me zeK, Fe ei aRC & AdamRL. Classi ica ion o hy oid ollicula lesions based on nuclea ex u e ea u es-lesion size ma e s. Cy ome y 2010 77 1101–1102. (h ps://doi.o g/10.1002/ cy o.a.20982) 14 LloydRV, Osamu aRY, KloppelG & RosaiJ. WHO Classi ica ion o Tumou s o Endoc ine O gans, 4 h ed. Lyon, F ance: IARC P ess, 2017. 16 C issmanJD, D ozdowiczS, JohnsonC & KiniSR. Fine needle aspi a ion diagnosis o hype plas ic and neoplas ic ollicula nodules o he hy oid – a mo phome ic s udy. Analy ical and Quan i a i e Cy ology and His ology 1991 13 321–328. 17 LohW-Y. Classi ica ion and eg ession ees. WIREs Da a Mining and Knowledge Disco e y 2011 1 14–23. (h ps://doi.o g/10.1002/widm.8) 18 GilJ & WuHS. Applica ion o image analysis o ana omic pa hology: eali ies and p omises. Cance In es iga ion 2003 21 950–959. (h ps:// doi.o g/10.1081/CNV-120025097) 19 Kayse K, HoshangSA, Me zeK, GoldmannT, Vollme E, RadziszowskiD, Kosje inaZ, Mi eskanda iM & Kayse G. Tex u e- and objec - ela ed au oma ed in o ma ion analysis in his ological s ill images o a ious o gans. Analy ical and Quan i a i e Cy ology and His ology 2008 30 323–335. 20 Mul aneL, RexhepajE, PenneyS, CallananJJ & Gallaghe WM. Au oma ed image analysis in his opa hology: a aluable ool in medical diagnos ics. Expe Re iew o Molecula Diagnos ics 2008 8 707–725. (h ps://doi.o g/10.1586/14737159.8.6.707) 21 NikonenkoAG & BozhokYM. Simple cumpu a ional echnique o quan i y nuclea shape asymme y. Cy ome y 2015 87 309–314. (h ps://doi.o g/10.1002/cy o.a.22612) 22 DoudkineA, MacaulayC, PoulinN & PalcicB. Nuclea ex u e measu emen s in image cy ome y. Pa hologica 1995 87 286–299. 23 Niki o o YE, See halaRR, TalliniG, BalochZW, BasoloF, ThompsonLD, Ba le aJA, WenigBM, Al GhuzlanA, KakudoK, e al. Nomencla u e e ision o encapsula ed ollicula a ian o papilla y hy oid ca cinoma: a pa adigm shi o educe o e ea men o indolen umo s. JAMA Oncology 2016 8 1023–1029. (h ps://doi. o g/10.1001/jamaoncol.2016.0386) 24 Male aF, MassaF, To eg ossaL, Du egonE, CasadeiGP, BasoloF, TalliniG, Volan eM, Niki o o YE & Papo iM. Cy ological ea u es o ‘nonin asi e ollicula hy oid neoplasm wi h papilla y- like nuclea ea u es’ and hei co ela ion wi h umo his ology. Human Pa hology 2016 54 134–142. (h ps://doi.o g/10.1016/j. humpa h.2016.03.014) 25 ShihSR, ChangYC, LiHY, LiauJY, LeeCY, ChenCM & ChangTC. P eope a i e p edic ion o papilla y hy oid ca cinoma p ognosis wi h he assis ance o compu e ized mo phome y o cy ology samples ob ained by ine needle aspi a ion: p elimina y epo . Head and Neck 2013 35 28–34. (h ps://doi.o g/10.1002/hed.22909) Recei ed in inal o m 20 June 2018 Accep ed 3 July 2018 Accep ed P ep in published online 4 July 2018 This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License. h ps://doi.o g/10.1530/EC-18-0237 h p://www.endoc ineconnec ions.o g © 2018 The au ho s Published by Bioscien i ica L d Downloaded om Bioscien i ica.com a 01/13/2020 05:40:14PM ia ee access