Ngiyabonga ngokuvakashela imvelo.com. Uhlobo lwesiphequluli osisebenzisayo lunokusekelwa kwe-CSS okukhawulelwe. Ngemiphumela emihle, sincoma ukusebenzisa inguqulo entsha yesiphequluli sakho (noma ukucisha imodi yokuhambisana ku-Internet Explorer). Okwamanje, ukuqinisekisa ukwesekwa okuqhubekayo, sibonisa isiza ngaphandle kwesitayela noma iJavascript.
Amazinyo abhekwa njengenkomba enembile kunazo zonke yobudala bomzimba womuntu futhi ajwayele ukusetshenziswa ekuhlolweni kweminyaka yobudala. Sihlose ukuqinisekisa ukulinganiselwa kweminyaka yemali esekelwe kwimibanweni ngokuqhathanisa ukunemba kokulinganisa kanye nokuhlukaniswa komkhawulo wonyaka we-18 ngezindlela zendabuko kanye nezilinganiso zeminyaka yemayini yedatha. Ingqikithi yemisebe yephaneli engama-2657 yaqoqwa kusuka kwizakhamizi zaseKorea nezamaJapan ezineminyaka engu-15 kuye kwengama-23. Bahlukaniswe ngesethi yokuqeqeshwa, ngalinye liqukethe ama-radiographs aseKorea angama-900, kanye nokuhlolwa kwangaphakathi okuqukethe ama-radiograph angama-857 aseJapan. Siqhathanise ukunemba kokuhlukaniswa kwezindlela zokuhlukaniswa kanye nokusebenza kahle kwezindlela zendabuko ngamasethi wokuhlola amamodeli wezimayini zedatha. Ukunemba kwendlela yendabuko kusethi yokuhlola kwangaphakathi kuphakeme kancane kunalokho kwemodeli yezimayini yedatha, futhi umehluko mncane (kusho iphutha eliphelele <0.21 iminyaka yesikwele <imizuzu engu-0.24). Ukusebenza kokuhlukaniswa kwe-cutoff yeminyaka eyi-18 nakho kuyafana phakathi kwezindlela zendabuko namamodeli wokumbiwa kwedatha. Ngakho-ke, izindlela zendabuko zingathathelwa indawo ngamamodeli wokumba idatha lapho wenza ukuhlolwa kweminyaka yobudala esebenzisa ukuvuthwa kwe-molars yesibili neyesithathu entlentshonalanga yaseKorea nabantu abadala.
Ukulinganiselwa kweminyaka yamazinyo kusetshenziswa kabanzi kumuthi we-forensic kanye ne-pediatric dentistry. Ikakhulu, ngenxa yokuhlobana okuphezulu phakathi kweminyaka yokulandelana kwezikhathi kanye nokuthuthukiswa kwamazinyo, ukuhlolwa kweminyaka kwezigaba zentuthuko yamazinyo kuyindlela ebalulekile yokuthola iminyaka yobudala yokuhlola iminyaka yezingane nentsha nentsha yentsha ),2,3. Kodwa-ke, kubantu abasha, kulinganisa iminyaka yamazinyo ngokususelwa ekuvuthweni kwamazinyo kuba nokulinganiselwa kwayo ngoba ukukhula kwamazinyo kucishe kuqedwe, ngaphandle kwezindunduma zesithathu. Injongo yezomthetho yokunquma iminyaka yentsha nentsha ukuhlinzeka ngokulinganisa okunembile kanye nobufakazi besayensi bokuthi bafinyelele eminyakeni yobudala obukhulu. Emthethweni wezomthetho wezomthetho wentsha kanye nabantu abadala abaseKorea, kulinganiselwa kusetshenziswa indlela kaLee, futhi kwabikezelwa umkhawulo ongokomthetho ngokususelwa kudatha ebikwe yidatha ebikwe ngu-Oh et al 5.
Ukufundwa komshini kuwuhlobo lobuhlakani bokufakelwa (AI) obufunda kaninginingi futhi kuhlukanise inani elikhulu ledatha, kuxazulula izinkinga ngeyakho, futhi kuqhuba uhlelo lwedatha. Ukufundwa komshini kungathola amaphethini afihlekile afihlekile kumavolumu amakhulu we-Data6. Ngokuphambene nalokho, izindlela zakudala, ezisebenza kanzima futhi ezidla isikhathi, zingase zibe nemikhawulo lapho zibhekene namanani amakhulu emininingwane eyinkimbinkimbi okunzima ukucubungula ngesandla7. Ngakho-ke, izifundo eziningi zenziwe muva nje zisebenzisa ubuchwepheshe bamuva bekhompyutha ukunciphisa amaphutha abantu futhi ngempumelelo inqubo yedatha ye-multidimensional ,,9,10,112,112 ,12. Ikakhulu, ukufundwa okujulile kusetshenziswe kabanzi ekuhlaziyweni kwezithombe zezokwelapha, kanye nezindlela ezahlukahlukene zokulinganisa iminyaka ngokuhlaziya ama-radiographs ngokuzenzakalela ukuthuthukisa ukunemba kanye nokusebenza kahle kweminyaka yobudala13,14,15,16,17,18,16,17,18,19,20 . Isibonelo, iHalabi et al 13 yathuthukisa i-algorithm yokufunda umshini ngokusekelwe kumanethiwekhi we-neural aural (CNN) ukuba alinganise iminyaka yamathambo esebenzisa ama-radiographs ezandla zezandla zezingane. Lolu cwaningo luhlongoza imodeli esebenza ngokufunda komshini ezithombeni zezokwelapha futhi ibonisa ukuthi lezi zindlela zingathuthukisa ukunemba kokuxilonga. I-LI et al14 yobudala obulinganiselwa kusuka ku-Pelvic X-Ray Izithombe zisebenzisa i-CNN ejulile futhi zizifanisa nemiphumela yokuhlehlisa usebenzisa isilinganiso sesigaba se-Ossification. Bathole ukuthi imodeli yokufunda ejulile ye-CNN ikhombisa ukusebenza okufanayo kweminyaka yobudala njengemodeli yokubuyiselwa kwendabuko. Ucwaningo lwe-Guo et al. [15] Kuhlolwe ukusebenza kokuhlukaniswa kweminyaka yobudala yobuchwepheshe be-CNN kususelwa kumazinyo ama-Orthophotos, kanti imiphumela yemodeli ye-CNN ifakazele ukuthi abantu babeka ukusebenza kwayo kokuhlukaniswa kweminyaka yobudala.
Izifundo eziningi ezilinganiselwa kubudala besebenzisa ukufundwa komshini Sebenzisa izindlela zokufunda ezijulile13,14,15,16,17,18,19,20. Ukulinganiselwa kweminyaka kususelwa ekufundeni okujulile kubikwa ukuthi kunembe kakhudlwana kunezindlela zendabuko. Kodwa-ke, le ndlela inikeza ithuba elincane ukwethula isisekelo sesayensi sokuqagela kweminyaka yobudala, njengezinkomba zeminyaka esetshenziswe ezilinganisweni. Kukhona nengxabano esemthethweni yokuthi ngubani oqhuba ukuhlolwa. Ngakho-ke, ukulinganiselwa kweminyaka okusekelwe ekufundeni okujulile kunzima ukwamukela ngeziphathimandla zezokuphatha nezesahlulelo. I-Data Mining (DM) iyindlela engatholanga bekulindelwe nje kuphela kodwa futhi imininingwane engalindelekile njengendlela yokuthola ukuxhumana okuwusizo phakathi kwamanani amakhulu e-Data6,21,22. Ukufunda ngomshini kuvame ukusetshenziswa kwimiyini yedatha, futhi zombili izimayini zedatha kanye nokufunda komshini Sebenzisa ama-algorithms afanayo okuthola amaphethini kudatha. Ukulinganiselwa kweminyaka usebenzisa ukuthuthukiswa kwamazinyo kususelwa ekuhlolweni komhloli wokuvuthwa kwamazinyo athambekile, futhi lokhu kuhlola kuvezwa njengesiteji sezinyo ngalinye eliqondiwe. I-DM ingasetshenziselwa ukuhlaziya ukuxhumeka phakathi kwesigaba sokuhlola samazinyo kanye nobudala bangempela futhi kunamandla okufaka esikhundleni sokuhlaziywa kwezibalo zendabuko. Ngakho-ke, uma sisebenzisa amasu we-DM ekulinganiseni kweminyaka yobudala, singasebenzisa ukufunda ngomshini ekulinganiseni iminyaka yobudala be-forensic ngaphandle kokukhathazeka ngesikweletu esingokomthetho. Izifundo ezimbalwa zokuqhathanisa zishicilelwe ngezindlela ezingenzeka ngezindlela zendabuko zendabuko ezisetshenziswa ekusebenzeni kwe-forensic kanye nezindlela ezenzelwe ama-ebm zokunquma iminyaka yamazinyo. UShen et al23 wakhombisa ukuthi imodeli ye-DM inembe kakhudlwana kunefomula yekhamera yendabuko. I-GALIBourg et al24 yafaka izindlela ezahlukahlukene zokubikezela ubudala ngokuya nge-demirdjian criterione25 kanti imiphumela iveze ukuthi izindlela ze-DM zaphazamisa izindlela ze-demirdjian kanye ne-willems ekulinganiseni iminyaka yenani labantu baseFrance.
Ukulinganisa iminyaka yamazinyo yentsha yaseKorea nabantu abadala, indlela kaLee 4 isetshenziswa kabanzi ekusebenzeni kwe-korea. Le ndlela isebenzisa ukuhlaziywa kwezibalo zendabuko (njengokubuyiselwa okuningi) ukuhlola ubudlelwano phakathi kwezifundo zaseKorea kanye nobudala obulandelana kwezikhathi. Kulolu cwaningo, izindlela zokulinganisa iminyaka ezitholakele zisebenzisa izindlela zendabuko zezibalo zichazwa ngokuthi "izindlela zendabuko." Indlela kaLee yindlela yendabuko, futhi ukunemba kwawo kuqinisekisiwe ngu-Oh et al. 5; Kodwa-ke, ukusebenza kokulinganisa kweminyaka kususelwa kwimodeli ye-DM ekusebenzeni kwe-koreas kwe-forensic kusangabazekayo. Umgomo wethu kwakuwukuqinisekisa ngokwesayensi usizo olungaba khona lokulinganisa kweminyaka yobudala ngokususelwa kumodeli ye-DM. Inhloso yalolu cwaningo (1) ukuqhathanisa ukunemba kwamamodeli amabili we-DM ekulinganiseni iminyaka yamazinyo futhi (2) ukuqhathanisa ukusebenza kwezigaba zamamodeli angama-7 dm eneminyaka eyi-18 nalezo ezitholakele ukuvuthwa kwezibalo zendabuko kwesibili kanye ne-molars yesithathu emihlathini yomibili.
Kusho ukuphambuka okujwayelekile kweminyaka yobudala yokulandelana kwezikhathi ngesiteji nohlobo lwamazinyo kukhonjisiwe online kuThebula Lesengezo S1 (Setha Lokuqeqeshwa), Ithebula Lokungezelela Langaphakathi (isethi yokuhlola yangaphandle). Amanani weKappa wokuthembela kwe-Intra- kanye nokwethenjwa kwe-InteralServer okutholakele kusuka kusethi yokuqeqeshwa kwakungu-0.951 no-0.947, ngokulandelana. Amanani we-P nezikhathi ezingama-95% zokuzethemba zamanani we-kappa zikhonjiswa kuthebula lesengezo le-inthanethi s4. Inani leKappa lahunyushwa ngokuthi "cishe liphelele", lihambisana nenqubo yeLandis neKoch26.
Uma uqhathanisa iphutha eliphelele elishoyo (i-mae), indlela yendabuko kancane ihlume imodeli ye-DM yabo bonke abesilisa nabesilisa kanye nokuhlolwa kwabesilisa kwangaphandle, ngaphandle kwe-Multilayer Perceptron (MLP). Umehluko phakathi kwemodeli yendabuko kanye nemodeli ye-DM kwisethi ye-MAE yangaphakathi ye-Mae kwakuyiminyaka engu-0.1-0-0.19 yabesilisa kanye ne-0.17-0.21 iminyaka yabesifazane. Ibhethri lokuhlola langaphandle, umehluko mncane (0.001-0.055 iminyaka yamadoda nezinkulungwane eziyi-0,05-0.09 zabesifazane). Ngokwengeziwe, impande isho iphutha lesikwele (i-RMSE) liphansi kancane kunendlela yendabuko, ngokuhluka okuncane (0.17-0.24, 0.03-0.07, 0.04-0.08 ngesethi yokuhlola yangaphandle). ). I-MLP ikhombisa ukusebenza okungcono kancane kune-ungqimba eyodwa ye-Perceptron (SLP), ngaphandle kokuthi kusethi yangaphandle yangaphandle yensikazi. Okwe-mae kanye ne-RMSE, izivivinyo zangaphandle zisetha izikolo eziphakeme kunokuhlolwa kwangaphakathi okubekwe kubo bonke abesilisa nabesilisa namamodeli. Yonke i-mae ne-rmse kuboniswa kuThebula 1 naku-1.
Mae kanye ne-rmse amamodeli wendabuko kanye nedatha yokubuyiselwa kwedatha. Kusho iphutha eliphelele mae, izimpande kusho iphutha lesikwele
Ukusebenza ngezigaba (nge-cutoff yeminyaka engu-18) yamamodeli wendabuko kanye ne-DM kukhonjiswe ngokuya ngokuzwela, ukucaciswa, inani elifanele lokubikezela (i-NPV), nendawo engaphansi kwendawo yokuqagela esebenzayo (Auroc) 27 (Ithebula 2, Umdwebo 2 nokwengeza umfanekiso 1 online). Ngokuya ngokuzwela kwebhethri lokuhlola langaphakathi, izindlela zendabuko zenziwa kangcono kakhulu phakathi kwabesilisa futhi zimbi phakathi kwabesifazane. Kodwa-ke, umehluko ekusebenzeni kokuhlukaniswa kwezindlela zendabuko kanye ne-SD yi-9.7% yamadoda (i-MLP) kanye nama-2.4% kuphela abesifazane (xgboost). Phakathi kwamamodeli we-DM, ukubuyiselwa imali (i-LR) kubonise ukuzwela okungcono kubo bobabili ubulili. Mayelana nokucaciswa kokuhlolwa kwangaphakathi kwesethi yangaphakathi, kwaqashelwa ukuthi amamodeli amane we-SD enza kahle kwabesilisa, kuyilapho imodeli yendabuko yenziwa kangcono kubantu besifazane. Umehluko ekusebenzeni kwezigaba kwabesilisa nabesifazane bangama-13.3% (MLP) no-13.1% (MLP), ngokulandelana, okubonisa ukuthi umehluko ekusebenzeni kokuhlukanisa phakathi kwamamodeli aphakathi kwamamodeli aphakathi kwamamodeli. Phakathi kwamamodeli we-DM, umshini we-vector osekelayo (SVM), amamodeli wesinqumo (i-DT), namamodeli wehlathi angahleliwe (ama-RF) awenzile okuhle kakhulu phakathi kwabesilisa, ngenkathi imodeli ye-LR eyenziwa kahle phakathi kwabesifazane. I-auroc yemodeli yendabuko kanye nawo wonke amamodeli we-SD ayekhulu kune-0.925 (k-ndia umakhelwane (knn) emadodeni), ekhombisa ukusebenza okuhle kwesampula eneminyaka engu-18. Ngokuhlolwa kwangaphandle kwesethi, kwaba nokuncipha ekusebenzeni kokuhlukaniswa ngokwemibandela yokuzwela, ukucaciswa kanye ne-auroc ngokuqhathaniswa nokuhlolwa kwangaphakathi. Ngaphezu kwalokho, umehluko wokuzwela nokucaciswa phakathi kokusebenza kokuhlukaniswa kwamamodeli amahle kakhulu futhi amabi asuka ku-10% kuya ku-25% futhi ube mkhulu kunomehluko ekuhlolweni kwangaphakathi.
Ukuzwela nokucaciswa kwamamodeli wokuhlelwa kwezimayini wezimayini kuqhathaniswa nezindlela zendabuko ezine-cutoff yeminyaka eyi-18. Umakhelwane oseduzane umakhelwane oseduzane, umshini we-SVM wokusekela umshini, umngcele we-LR, isihlahla sesinqumo se-DT, ihlathi elingenamqondo, i-XGB xgboost, i-MLP Multilayer Perceptron, indlela yendabuko ye-CM.
Isinyathelo sokuqala kulolu cwaningo bekuwukuqhathanisa ukunemba kwezilinganiso zeminyaka yamazinyo etholakala kumamodeli we-DM ayisikhombisa nalezo ezitholakele zisebenzisa ubuciko bendabuko. UMae noRmse bahlolwe emabekweni kwangaphakathi kwazo zombili izicansi, futhi umehluko phakathi kwendlela yendabuko kanye nemodeli ye-DM kuye kwayi-7 kuye kwezingama-77 nge-MMSSE. Yize indlela yendabuko inembe kakhudlwana kulolu cwaningo, kunzima ukuphetha ukuthi umehluko omncane unokubaluleka kokwenza emtholampilo noma okusebenzayo. Le miphumela ikhombisa ukuthi ukunemba kokulinganisa kweminyaka yamazinyo kusetshenziswa imodeli ye-DM kucishe kufane nangendlela yendabuko. Ukuqhathanisa okuqondile nemiphumela evela ezifundweni ezedlule kunzima ngoba akukho cwaningo okufane ngokunemba kwamamodeli we-DM ngezindlela zendabuko zezibalo usebenzisa izindlela ezifanayo zokuqopha amazinyo afinyelela kulolu cwaningo. I-GALIBourg et al24 iqhathaniswa mae kanye ne-RMSE phakathi kwezindlela ezimbili zendabuko (indlela yeDemirji Method25 neWillems Medical29) namamodeli we-10 DM kubantu baseFrance abaneminyaka emibili ubudala abaneminyaka engama-2 kuya kwayi-24. Babike ukuthi wonke amamodeli we-DM anembe kakhulu kunezindlela zendabuko, ngokuhlukahluka kwe-0.20 kanye ne-0.38 iminyaka engu-0 Ukungafani phakathi kwemodeli ye-SD kanye nezindlela zendabuko eziboniswe esifundweni se-HELIBOURG kubhekisisa imibiko eminingi engamakhulu amathathu,31,32,33 ukuthi indlela ye-demirdjian ayilinganisela kahle iminyaka yobudala yamazinyo kubantu abafundwayo lapho isifundo sasisekelwa khona. kulolu cwaningo. UTi et al 34 wasebenzisa i-MLP algorithm ukubikezela iminyaka yezinyo kusuka ezithombeni ezingama-1636 Chinese orthodontic futhi waqhathanisa ukunemba kwayo nemiphumela yendlela yeDemirji neWillems. Babike ukuthi i-MLP inokunemba okuphakeme kunezindlela zendabuko. Umehluko phakathi kwendlela ye-demirdjian nendlela yendabuko yiminyaka engu-0 0.32, futhi indlela ye-willems yiminyaka engu-0.28, efana nemiphumela yesifundo samanje. Imiphumela yalezi zifundo ezedlule24,34 iyahambisana nemiphumela yesifundo samanje, kanye nokunemba kokulinganisa kweminyaka yemodeli ye-DM kanye nendlela yendabuko kuyafana. Kodwa-ke, ngokuya ngemiphumela eyethulwe, singaphetha ngokucophelela ngokuthi ukusetshenziswa kwamamodeli we-DM ukulinganisa iminyaka ekhona kungathatha isikhundla sezindlela ezikhona ngenxa yokuntuleka kwezifundo zokuqhathanisa nezibhekisayo. Izifundo ezilandelwayo ezisebenzisa amasampula amakhulu ziyadingeka ukuqinisekisa imiphumela etholakele kulolu cwaningo.
Phakathi kwezifundo ezivivinya ukunemba kwe-SD ekulinganiseni iminyaka yamazinyo, abanye bakhombisa ukunemba okuphezulu kunokutadisha kwethu. I-Stepanovsky et al 35 isetshenziswe amamodeli angama-22 e-SD kuma-radiographs we-panoramic wezakhamizi ezingama-976 zaseCzech ezineminyaka engu-276 kuya kwengama-20,5 kuye kwangama-20,5 futhi zihlole ukunemba kwemodeli ngayinye. Baye bahlola ukuthuthukiswa kwamazinyo angapheli aphezulu ayi-16 angenayo asebenzisa izindlela zokuhlukaniswa ezihlongozwe yiMoorrees et al 36. Ama-mae aqala ukusuka eminyakeni engu-0.64 kuye kwayi-0.94 kanye nama-RMSE asuka eminyakeni engu-0.85 kuya ku-1.27, anembe kakhudlwana kunamamodeli amabili we-DM asetshenziswe kulolu cwaningo. UShen et al23 wasebenzisa indlela yeCameriere ukukala iminyaka yamazinyo ayisikhombisa yaphakade e-Eastern Chinese Izakhamizi ezineminyaka engu-5 kuya kwengu-13 futhi iqhathanise iminyaka ecatshangelwe isebenzisa ukuhlelwa kabusha okuqondile, i-SVM ne-RF. Bakhombisa ukuthi wonke amamodeli amathathu we-DM anokunemba okuphezulu uma kuqhathaniswa nefomula yendabuko yeCameriere. I-mae noRmse esifundweni seShen bebephansi kunaleyo esemodeni ye-DM kulolu cwaningo. Ukucaciswa okwengeziwe kwezifundo nguStepanovsky et al. 35 noShen et al. 23 Kungenzeka kube ngenxa yokufakwa kwezihloko ezincane kumasampula awo okufunda. Ngenxa yokuthi ukulinganiselwa kweminyaka kwabahlanganyeli abanamazinyo athuthukayo baba banembe ngokwengeziwe njengoba inani lamazinyo likhuphuka ngesikhathi sokuthuthuka kwamazinyo, ukunemba kwendlela yokulinganisa yobudala kungahle kubekwe phansi lapho kubambe iqhaza ekuhlolweni. Ngokwengeziwe, iphutha le-MLP ekulinganiseni kweminyaka lincane kakhulu kune-SLP's, okusho ukuthi i-MLP inembe kakhulu kune-SLP. I-MLP ibhekwa njengengcono kakhulu ekulinganiselweni kweminyaka yobudala, mhlawumbe ngenxa yezendlalelo ezifihliwe ku-MLP38. Kodwa-ke, kukhona okuhlukile kwesampula yangaphandle yabesifazane (SLP 1.45, MLP 1.49). Ukuthola ukuthi i-MLP inembe kakhudlwana kune-SLP ekuhlolweni kweminyaka idinga izifundo ezengeziwe zokuthola kabusha.
Ukusebenza kwehlukaniswa kwemodeli ye-DM kanye nendlela yendabuko ngombundu weminyaka engu-18 nakho kuqhathaniswa. Zonke izinhlobo ze-SD ezihloliwe kanye nezindlela zendabuko ezisethiwe zokuhlola zangaphakathi zikhombise amazinga amukelekile okubandlululwa kwesampula eneminyaka engu-18 ubudala. Ukuzwela kwabesilisa nabesifazane bekukhulu kune-87.7% no-94.9%, ngokulandelana, futhi ukucaciswa bekukhulu kune-89.3% no-84.7%. I-auroc yazo zonke izinhlobo ezihloliwe nayo idlula u-0.925. Ngokwazi okuhle kolwazi lwethu, akukho cwaningo oluhlole ukusebenza kwemodeli ye-DM ukuhlukaniswa kweminyaka engu-18 ngokususelwa ekuvuthweni kwamazinyo. Singaqhathanisa imiphumela yalolu cwaningo ngokusebenza kokuhlukaniswa kwamamodeli wokufunda ajulile kuma-radiographs panoramic. I-Guo et al.15 Ibalwe ukusebenza kokuhlukaniswa kwemodeli yokufunda esekwe e-CNN kanye nendlela ebhaliwe ngokususelwa kwindlela yeDemirji yomkhawulo othile weminyaka. Ukuzwela nokucaciswa kwendlela ebhaliwe kwakungu-87.7% no-95,5%, ngokulandelana, kanye nokuzwela nokucaciswa kwemodeli ye-CNN kwedlula ama-89.2% no-86.6%, ngokulandelana. Baphetha ngokuthi amamodeli wokufunda ajulile angangena esikhundleni noma ukuhlolwa kwemalukwe ngaphandle kokuhlukanisa imibundu yobudala. Imiphumela yalolu cwaningo ikhombise ukusebenza okufanayo kokuhlukaniswa; Kukholelwa ukuthi ukuhlukaniswa kusetshenziswa amamodeli we-DM kungathatha isikhundla sezindlela zendabuko zezibalo zokulinganisa kweminyaka yobudala. Phakathi kwamamodeli, i-DM LR yayiyimodeli engcono kakhulu ngokuya ngokuzwela kwesampula yabesilisa nokuzwela nokucaciswa kwesampula yabesifazane. I-LR ibekwe okwesibili ngokucaciswa kwabesilisa. Ngaphezu kwalokho, i-LR ibhekwa njengenye yamamodeli we-DM35 asebenziseka kalula asebenziseka kalula futhi ayinkimbinkimbi futhi enzima ukuyicubungula. Ngokusekelwe kule miphumela, i-LR yayibhekwa njengemodeli yokuhlukaniswa okuhle kakhulu kwe-cutoff yabantwana abaneminyaka engu-18 ubudala kubantu baseKorea.
Sekukonke, ukunemba kokulinganisa kweminyaka noma ukusebenza kokuhlukaniswa kokuhlolwa kwangaphandle kwesethi yangaphandle kwakumpofu noma kuphansi uma kuqhathaniswa nemiphumela kusethi yangaphakathi yokuhlola. Eminye imibiko ikhombisa ukuthi ukunemba kokuhlukaniswa kohlukaniso kuncipha lapho kulinganiselwa iminyaka yobudala obususelwa kubantu baseKorea kusetshenziswa kubantu baseJapan5,39, futhi kwatholakala indlela efanayo esifundweni samanje. Lo mkhuba wokuwohloka ubuye waqashelwa kwimodeli ye-DM. Ngakho-ke, ukukala ngokunembile iminyaka yobudala, noma ngabe usebenzisa i-DM inqubo yokuhlaziya, izindlela ezithathwe kwimininingwane yendabuko, njengezindlela zendabuko, kufanele zikhethwe. Njengoba akucaci ukuthi ngabe amamodeli wokufunda ajulile angabonisa ukuthambekela okufanayo, izifundo ngokuqhathanisa ukunemba nokusebenza kwendabuko, amamodeli we-DM, namamodeli wokufunda ajulile kumasampula afanayo adinga ukunquma nalokhu kubonakala ngokobuhlanga. ukuhlolwa.
Sibonisa ukuthi izindlela zendabuko zingathathelwa indawo ukulinganiselwa kweminyaka kususelwa kwimodeli ye-DM ekuzijwayezeni kokulinganisa iminyaka yobudala eKorea. Sithole nokuthi kungenzeka ukuthi kusetshenziswe umshini wokufunda ngomshini wokuhlola iminyaka yobudala. Kodwa-ke, kunemikhawulo ecacile, efana nenombolo enganele yabahlanganyeli kulolu cwaningo ukuze banqume imiphumela, kanye nokuntuleka kwezifundo ezedlule ukuqhathanisa nokuqinisekisa imiphumela yalolu cwaningo. Ngokuzayo, izifundo ze-DM kufanele zenziwe ngamanani amakhulu amasampula kanye nezindawo ezahlukahlukene ukuthuthukisa ukusebenza kwalo okusebenzayo uma kuqhathaniswa nezindlela zendabuko. Ukuze uqinisekise amandla okusebenzisa ubuhlakani bokufakelwa ukulinganisa iminyaka eminingi, izifundo zesikhathi esizayo ziyadingeka ukuqhathanisa ukunemba nokusebenza kwe-DM kanye namamodeli wokufunda ajulile ngezindlela zendabuko kumasampula endabuko.
Ucwaningo lusebenzise izithombe ezingama-2,657 ze-orthographic eziqoqwe kusuka kubantu abadala baseKorea nabaseJapan abaneminyaka engu-15 kuye kwengama-23. Ama-radiographs aseKorea ahlukaniswe ngamasethi wokuqeqeshwa angama-900 (19.42 ± 2.65 iminyaka) kanye nama-900 wokuhlola kwangaphakathi (19.52 ± iminyaka engu-1900). Isethi yokuqeqeshwa yaqoqwa esikhungweni esisodwa (Seoul St Mary's Hospital), kanye nesethi yokuhlola uqobo yaqoqwa ezikhungweni ezimbili (isibhedlela saseSeoul National University Hospital kanye nesibhedlela saseYonsun University Dental Gospital). Siphinde saqoqa ama-radiographs angama-857 avela kwenye idatha esekelwe kwizinombolo (itate Medical University, eJapan) ngokuhlolwa kwangaphandle. Imisebe yemisebe yezihloko zaseJapan (19.31 ± 2.60 iminyaka) yakhelwe njengohlolo lwangaphandle olusethiwe. Idatha yaqoqwa ngokubuyiselwa emuva ukuze ihlaziye izigaba zokuthuthuka kwamazinyo kuma-radiographs we-panoramic athathwe ngesikhathi sokwelashwa kwamazinyo. Yonke imininingwane eqoqwe yayingaziwa ngaphandle kobulili, usuku lokuzalwa nosuku lwe-radiograph. Ukufakwa kanye nezindlela zokufakwa ngaphandle kwafana nezifundo ezishicilelwe ngaphambili ezishicilelwe 4, 5. Iminyaka yangempela yesampula yabalwa ngokususa usuku lokuzalwa kusukela ngosuku iRadiograph yathathwe ngalo. Iqembu lesampula lahlukaniswa ngamaqembu aneminyaka eyisishiyagalolunye. I-Age kanye nokusatshalaliswa kocansi kuboniswa kuThebula 3 Lolu cwaningo lwenziwe ngokuya ngesimemezelo seHelsinki futhi savunywa yiBhodi Yokubuyekezwa Kwezikhungo (i-IRB) yeSeoul St. Ngenxa yokwakheka okubusayo kwalolu cwaningo, imvume enolwazi ayitholakalanga kuzo zonke iziguli ukuhlolwa kwe-radiographic ngezinhloso zokwelapha. Seoul Korea University St. Mary's Hospital (i-IRB) wasusa isidingo semvume enolwazi.
Izigaba zentuthuko ze-Bimaxillary yesibili nezesithathu zihlolwe ngokuya nge-DemirCan Criteria25. Kukhethwe izinyo elilodwa kuphela uma kutholakala uhlobo lwezinyo olufanayo ngakwesobunxele nakwesokudla komhlathi ngamunye. Uma amazinyo anobungozi kuzo zombili izinhlangothi ayesezigabeni ezahlukahlukene zentuthuko, izinyo elinentuthuko ephansi likhethiwe ukuze ziphendule ngokungaqiniseki ngobudala obulinganiselwayo. Ama-radiographs ayikhulu akhethwe ngokungahleliwe avela kusethi yokuqeqeshwa atholwe ababukele ababili abanolwazi ukuze bahlole ukuthembeka okuhlobene ngemuva kokutholwa ukuthola isigaba sokuvuthwa kwamazinyo. Ukwethembeka kwe-IntraoBserver kuhlolwe kabili ngezikhathi ezinezinyanga ezintathu ngombukeli oyinhloko.
Isigaba sobulili kanye nentuthuko yesivumelwano sesibili nesithathu somhlathi ngamunye osendaweni yokuqeqeshwa balinganiselwa ngumbukeli oyinhloko abaqeqeshelwe amamodeli ahlukene we-DM, futhi iminyaka yangempela yahlelwa njengenani eliqondisiwe. Amamodeli we-SLP ne-MLP, asetshenziswa kabanzi ekufundeni komshini, ahlolwe ngokumelene nama-algorithms aphindaphindwe. Imodeli ye-DM ihlanganisa imisebenzi eqondile isebenzisa izigaba zentuthuko zamazinyo amane futhi ihlanganisa le mininingwane ukulinganisa iminyaka yobudala. I-SLP iyinethiwekhi elula ye-neural futhi ayiqukethe izingqimba ezifihliwe. I-SLP isebenza ngokusekelwe ekudluliseleni kwe-Threshold phakathi kwezindawo. Imodeli ye-SLP ku-Regression ifana ngokwezibalo Ngokungafani nemodeli ye-SLP, imodeli ye-MLP inezingqimba eziningi ezifihliwe ezinemisebenzi ye-nonlinear activation imisebenzi. Izivivinyo zethu zisebenzise ungqimba olufihliwe olunezindawo ezingama-20 ezifihliwe kuphela ngemisebenzi ye-nonlinear activation. Sebenzisa i-gradient fear njengendlela yokwenza kahle kanye ne-mae kanye ne-rmse njengomsebenzi wokulahleka ukuqeqesha imodeli yethu yokufunda yomshini. Imodeli yokuthola kabusha etholakele etholakele yasetshenziswa emabekweni okuhlola kwangaphakathi nangaphandle kanye neminyaka yamazinyo kulinganiselwa.
I-algorithm yehlukaniso yathuthukiswa esebenzisa ukuvuthwa kwamazinyo amane ekuqeqesheni ukubhekisisa ukubikezela ukuthi ngabe isampula ineminyaka eyi-18 noma cha. Ukwakha imodeli, sathola ukufundwa kwemishini eyisikhombisa yokufunda ama-algorithms6,43: (1) lr, (2) knn, (4) svm, (4) dt, (6) dt, (6) rup . I-LR ingenye ye-algorithms444 esetshenziswa kakhulu kunazo zonke. Kuyi-algorithm egadiwe yokufunda esebenzisa i-regression ukubikezela amathuba okuba idatha yesigaba esithile ukusuka ku-0 kuye ku-1 futhi ihlukanise imininingwane efana nesigaba esiseduze kwalokhu okusekelwe kulokhu okungenzeka; isetshenziselwa ikakhulukazi ukuhlukaniswa kanambambili. I-KNN ingenye yezinto ezilula zokufunda umshini algorithms45. Uma unikezwa idatha yokufaka entsha, uthola idatha ye-K eduze kwesethi ekhona bese uyihlukanisa ekilasini elinemvamisa eliphakeme kakhulu. Sibeka i-3 ngenani labomakhelwane babheka (k). I-SVM i-algorithm eyandisa ibanga phakathi kwamakilasi amabili ngokusebenzisa umsebenzi we-kernel ukwandisa isikhala somugqa endaweni engekho emthethweni obizwa nge-field46. Kule modeli, sisebenzisa i-bias = 1, amandla = 1, kanye ne-gamma = 1 njengama-hyperparemeters we-polynomial kernel. I-DT isetshenziswe emikhakheni ehlukahlukene njenge-algorithm yokuhlukanisa yonke imininingwane ebekwe emaqenjini ambalwa ngokumele imithetho yezinqumo ngesakhiwo sesihlahla47. Imodeli ilungiselelwe ngenani eliphansi lamarekhodi nge-node ye-2 futhi isebenzisa inkomba ye-Gini njengesilinganiso sekhwalithi. I-RF iyindlela ye-Encleng ehlanganisa ama-DTS amaningi ukuze athuthukise ukusebenza usebenzisa indlela yokuhlanganisa i-Bootsstrap ekhiqiza i-classifier ebuthakathaka yesampula ngalinye ngamasampula afanayo afanayo amahlandla amaningi avela kuDamplet48. Sisebenzise izihlahla eziyi-100, ukujula kwezihlahla eziyi-10, usayizi we-node ongu-1, kanye ne-Gini Admixiture Index njengezindlela zokwahlukana kweminwe. Ukuhlukaniswa kwedatha entsha kunqunywa yivoti eliningi. I-XGBoost iyi-algorithm ehlanganisa amasu okukhulisa usebenzisa indlela ethatha idatha yokuqeqesha iphutha phakathi kwamanani wangempela futhi abikezelwe ngephutha usebenzisa i-gradient49. Kuyi-algorithm esetshenziswa kabanzi ngenxa yokusebenza kwayo okuhle kanye nokusebenza kwezinsizakusebenza, kanye nokwethenjwa okuphezulu njengomsebenzi wokulungiswa ngokweqile. Imodeli ifakwe amasondo okusekelwa angama-400. I-MLP iyinethiwekhi ye-neural lapho eyodwa noma ngaphezulu ye-PeCCEPTRONS yakha izingqimba eziningi ngezendlalelo esisodwa noma eziningi ezifihliwe phakathi kokufakwayo nokukhipha i-autler38. Usebenzisa lokhu, ungenza ukuhlukaniswa okungekho emthethweni lapho ufaka ungqimba wokufaka bese uthola inani lemiphumela, inani elibikezelwe liqhathaniswa nenani langempela lemiphumela futhi iphutha lisatshalaliswa emuva. Sidale ungqimba olufihliwe ngama-neurons angama-20 afihliwe engxenyeni ngayinye. Imodeli ngayinye esathuthukile sayisetshenziselwa amasethi angaphakathi nangaphandle ukuhlola ukusebenza kokuhlukaniswa ngokubala ukuzwela, ukucaciswa, i-ppv, ne-NPV, ne-auroc. Ukuzwela kuchazwa njengokulinganiswa kwesampula okulinganiselwa kuneminyaka engu-18 noma ngaphezulu ku-FAMET ebaluleke kakhulu ukuba yiminyaka eyi-18 noma ngaphezulu. Ukucaciswa yingxenye yamasampula amasampula aneminyaka engaphansi kweminyaka eyi-18 nalokho okulinganiselwa kungaphansi kweminyaka eyi-18.
Izigaba zamazinyo ezihlolwe kusethi yokuqeqeshwa zaziguqulwa zaba yizigaba zamanani zokuhlaziywa kwezibalo. I-Multivariate Linear ne-Regression yezenziwe kahle yenziwa ukuthuthukisa amamodeli wokuqagela kulowo ocansini ngakunye kanye nokuthola kabusha amafomethi wokuphinda abuyiselwe ukulinganisa iminyaka yobudala. Sisebenzise lezi zindlela zokulinganisa iminyaka yezinyo yezinyo zazo zombili izivivinyo zokuhlola zangaphakathi nangaphandle. Ithebula 4 libonisa ukuphelela kanye namamodeli wokuhlukaniswa asetshenziswe kulolu cwaningo.
Ukwethembeka kwe-Intra- kanye ne-InteralServer Ukuhlola ukunemba kwamamodeli we-DM kanye namamodeli wendabuko wendabuko, sabala ama-mae kanye ne-RMSE sisebenzisa iminyaka yobudala obulinganisiwe neyangaphandle yesethi yokuhlola yangaphakathi nangaphandle. Lawa maphutha ajwayele ukusetshenziselwa ukuhlola ukunemba kwezibikezelo zemodeli. Iphutha elincane, kuphakama ukunemba kwesimo sezulu24. Qhathanisa ama-mae kanye ne-rmse yamasethi wokuhlola wangaphakathi nangaphandle abaliwe kusetshenziswa i-DM nokubuyiselwa kwendabuko. Ukusebenza kokuhlukaniswa kwe-cutoff yeminyaka eyi-18 kwizibalo zendabuko kwahlolwa kusetshenziswa itafula eliphuthumayo le-2 × 2. Ukuzwela okubaliwe, ukucaciswa, i-PPV, i-NPV, ne-auroc yesethi yokuhlola kuqhathaniswa namanani alinganisiwe wemodeli yokuhlukaniswa kwe-DM. Idatha ivezwa njenge-adve ± ukuphambuka okujwayelekile noma inombolo (%) kuya ngezimpawu zedatha. Amanani e-P amanani amabili <0.05 abhekwa njengento ebalulekile ngokwezibalo. Konke ukuhlaziya kwezibalo okujwayelekile kwenziwa kusetshenziswa i-SAS version 9.4 (SAS Institute, Cary, NC). Imodeli yokuhlehlisa i-DM yenziwa ePython isebenzisa iKeras50 2.2.4 I-Backlend neTensorFlow51 1.8.0 ngokuqondile imisebenzi yezibalo. Imodeli yokuhlukaniswa kwe-DM yenziwa eWaikato Ulwazi Lokuhlaziywa Kolwazi kanye neKonstanz Information Miner (Knime) 4.6.152 Ipulatifomu yokuhlaziya.
Ababhali bayavuma ukuthi idatha exhasa iziphetho zocwaningo ingatholakala esihlokweni nasezintweni ezengeziwe. Ama-dataset akhiqizwa futhi / noma ahlaziywe ngesikhathi sokufunda ayatholakala kumlobi ohambelana nesicelo esifanele.
Ritz-timme, s. et al. Ukuhlolwa kweminyaka: Isimo sobuciko ukuze bahlangabezane nezidingo ezithile zokuzijwayeza kwe-forensic. ama-International. J. Ezomthetho Medicine. 113, 129-136 (2000).
I-Schmeling, a. Ama-forensics. umuthi. I-Pathology. 1, 239-246 (2005).
Pan, J. et al. Indlela elungisiwe yokuhlola iminyaka yamazinyo yezingane ezineminyaka engu-5 kuya kwayi-16 empumalanga yeChina. Clinical. Inhlolovo yomlomo. 25, 3463-3474 (2021).
I-Lee, SS njll. ama-International. J. Ezomthetho Medicine. 124, 659-665 (2010).
O, S., Kumagai, a. Plos One 17, E0271247 (2022).
UKim, jy, et al. Ukuhlaziywa kwedatha okusekelwe ekufundeni komshini okusekelwe ekufundeni kwedatha kungabikezela imiphumela yokwelashwa yokuhlinzwa yokulala ezigulini no-Osa. Isayensi. Bika 11, 14911 (2021).
Han, M. et al. Ukulinganiswa kweminyaka okunembile kusuka ekufundeni komshini noma ngaphandle kokungenelela komuntu? ama-International. J. Ezomthetho Medicine. 136, 821-831 (2022).
Khan, S. noShaheen, M. Kusuka kwimithombo yedatha kuya kwimiyini yedatha. J.Information. Isayensi. https://doi.org/10.1177/01655515211030872 (2021).
Khan, S. noShaheen, uM. Winrule: I-algorithm yokuqala ye-algorithm ye-Association Rule mining. J.Information. Isayensi. https://doi.org/10.1177/01655515221108695 (2022).
Shaheen M. no-Abdullah U. Bala. Matt. Qhubeka. 68, 3305-3322 (2021).
UMuhammad M., uRhman Z., Shaheen M., Khan M. noHabib M. Deep M. Deep Peep Okusekelwe Semantic Ukufana Ukuvela Kukuthola Idatha Yombhalo. yazisa. Ubuchwepheshe. ukulawula. https://doi.org/10.5755/j01.itc.49.4.27118 (2020).
Tabish, M., Tanoli, Z., noShahin, uM. Uhlelo lokubona umsebenzi kumavidiyo wezemidlalo. multimedia. Izicelo zamathuluzi https://doi.org/10.1007/s110422-021-10519-6 (2021).
Halabi, SS et al. Inselelo yokufunda yomshini we-RSNA e-Pediatric Bone Age. I-Radiology 290, 498-503 (2019).
I-LI, Y. et al. Ukulinganiselwa kweminyaka yobudala yobudala kusuka ku-pelvic x-ray usebenzisa ukufunda okujulile. I-Euro. imisebe. 29, 2322-2329 (2019).
Guo, YC, et al. Ukuhlukaniswa kweminyaka okunembile kusetshenziswa izindlela zezandla kanye namanethiwekhi ajulile we-neural neural avela ezithombeni ze-Orthographic Projection. ama-International. J. Ezomthetho Medicine. 135, 1589-1597 (2021).
U-Alabama Dalora et al. Ukulinganiselwa kweminyaka yobudala usebenzisa izindlela ezahlukahlukene zokufunda umshini: Ukubuyekezwa kwezincwadi okuhlelekile kanye nokuhlaziywa kwe-meta. Plos One 14, E0220242 (2019).
UDu, H., LI, G., CHEng, K., no-Yang, uJ. ama-International. J. Ezomthetho Medicine. 136, 811-819 (2022).
UKim S., Lee YH, NOH YK, Park FK kanye ne-OH KS Ukunquma Amaqembu Eminyaka Yabantu Abaphilayo Esebenzisa Izithombe Ezisekelwe Kwezobunhloli Zokufakelwa Kwemishini Yokuqala. Isayensi. Bika 11, 1073 (2021).
Stern, D., Payer, C., Giuliani, N., kanye ne-Urschler, M. I-othomathikhi yokulinganisa iminyaka kanye nokuhlukaniswa kweminyaka eminingi kusuka kudatha ye-MRIRIVARIET. Ieee J. biomed. Izaziso zezempilo. 23, 1392-1403 (2019).
UCheng, Q., UGe, ge, z., uDu, H. noLi, ukulinganiselwa kweminyaka yobudala ngokususelwa ku-3D pulp Chamber gembortion of molars yokuqala ngokuhlanganisa isethi yokufunda ejulile ngokuhlanganisa amasethi wokufunda ajulile ngokuhlanganisa amasethi wokufunda ajulile ngokuhlanganisa amasethi wokufunda ajulile ngokuhlanganisa amasethi wokufunda ajulile. ama-International. J. Ezomthetho Medicine. 135, 365-373 (2021).
I-Wu, wt, et al. Idatha yokumbiwa kwedatha enkulu yemitholampilo: Imininingwane ejwayelekile, izinyathelo, kanye nezindlela zeZindlela. Umhlaba. umuthi. izinsiza. 8, 44 (2021).
U-Yang, J. et al. Isingeniso kumininingwane yezokwelapha kanye nobuchwepheshe bezimayini zedatha kwi-Big Data Era. J. AVID. Umuthi oyisisekelo. 13, 57-69 (2020).
UShen, S. et al. Indlela kaCamerer yokulinganisa izinyo ubudala usebenzisa umshini wokufunda. I-BMC Oral Health 21, 641 (2021).
UGalliburg A. et al. Ukuqhathaniswa kwezindlela ezihlukile zokufunda umshini wokuqagela iminyaka yamazinyo usebenzisa indlela yesitayela se-demirdjian. ama-International. J. Ezomthetho Medicine. 135, 665-675 (2021).
UDemirdjian, A., Goldstein, H. noTanner, JM uhlelo olusha lokuhlola iminyaka yamazinyo. snort. Biology. 45, 211-2227 (1973).
I-Landis, JR, kanye noKoch, i-GG izindlela zesivumelwano sokubheka kwimininingwane yesigaba. Biometric 33, 159-174 (1977).
I-Bhattacharjee S, Prakash D, Kim C, Kim HK kanye neChoi HK. Ukuhlaziywa kwemibhalo, kwe-morphological kanye nezibalo zezinhlangothi ezimbili ze-magnetic resonance imaging usebenzisa amasu wokufakelwa kwezobunhloli wokwehlukanisa ama-tumors asemqoka. Imininingwane yezempilo. izinsiza. https://doi.org/10.4258/Hir.2022.28.1.46 (2022).
Isikhathi sePosi: Jan-04-2024