Siyabonga ngokuvakashela i-Nature.com.Inguqulo yesiphequluli oyisebenzisayo inosekelo olukhawulelwe lwe-CSS.Ukuze uthole imiphumela engcono kakhulu, sincoma ukuthi usebenzise inguqulo entsha yesiphequluli sakho (noma uvale imodi yokuvumelana ku-Internet Explorer).Okwamanje, ukuze siqinisekise ukwesekwa okuqhubekayo, sibonisa isayithi ngaphandle kwesitayela noma i-JavaScript.
Amazinyo abhekwa njengenkomba enembe kakhulu yobudala bomzimba womuntu futhi avame ukusetshenziswa ekuhloleni iminyaka ye-forensic.Sihlose ukuqinisekisa izilinganiso zeminyaka yamazinyo ezisekelwe ezimayini ngokuqhathanisa ukunemba kwesilinganiso nokusebenza ngokwezigaba komkhawulo weminyaka engu-18 nezindlela ezivamile kanye nezilinganiso zeminyaka yobudala ezisuselwe ekumbiweni kwedatha.Kwaqoqwa ama-panoramic radiographs angama-2657 ezakhamizini zaseKorea naseJapan ezineminyaka eyi-15 kuye kwengama-23.Zahlukaniswa zaba isethi yokuqeqeshwa, ngayinye iqukethe ama-radiographs e-Korean angu-900, kanye nesethi yokuhlola yangaphakathi equkethe ama-radiograph aseJapane angu-857.Siqhathanise ukunemba kwezigaba nokusebenza kahle kwezindlela zendabuko namasethi okuhlola amamodeli okumbiwa kwedatha.Ukunemba kwendlela evamile kusethi yokuhlola yangaphakathi iphakeme kancane kunaleyo yemodeli yezimayini yedatha, futhi umehluko mncane (kusho iphutha eliphelele
Ukufunda ngomshini kuwuhlobo lobuhlakani bokwenziwa (AI) olufunda ngokuphindaphindiwe futhi luhlukanise inani elikhulu ledatha, luxazulule izinkinga ngokwalo, futhi lushayele ukuhlelwa kwedatha.Ukufunda ngomshini kungathola amaphethini afihliwe awusizo kumthamo omkhulu wedatha6.Ngokuphambene, izindlela zakudala, ezisebenza kanzima futhi ezidla isikhathi, zingase zibe nemikhawulo lapho zisebenzelana nenani elikhulu ledatha eyinkimbinkimbi okunzima ukuyicubungula ngesandla7.Ngakho-ke, ucwaningo oluningi luye lwenziwa muva nje kusetshenziswa ubuchwepheshe bamuva bekhompiyutha ukuze kuncishiswe amaphutha abantu futhi kucutshungulwe kahle idatha ye-multidimensional8,9,10,11,12.Ikakhulukazi, ukufunda okujulile kuye kwasetshenziswa kabanzi ekuhlaziyeni izithombe zezokwelapha, futhi izindlela ezihlukahlukene zokulinganisa ubudala ngokuhlaziya ngokuzenzakalelayo ama-radiographs kuye kwabikwa ukuze kuthuthukiswe ukunemba nokusebenza kahle kokulinganisa kweminyaka13,14,15,16,17,18,19,20 .Isibonelo, u-Halabi et al 13 uthuthukise i-algorithm yokufunda yomshini esuselwe kumanethiwekhi e-convolutional neural (CNN) ukuze alinganisele ubudala bohlaka lwamathambo esebenzisa ama-radiographs ezandla zezingane.Lolu cwaningo luphakamisa imodeli esebenzisa umshini wokufunda ezithombeni zezokwelapha futhi lubonise ukuthi lezi zindlela zingathuthukisa ukunemba kokuxilonga.U-Li et al14 ulinganisela ubudala kusukela ezithombeni ze-X-ray ze-pelvic esebenzisa i-CNN yokufunda ejulile futhi waqhathanisa nemiphumela yokuhlehla kusetshenziswa ukulinganisa kwesiteji se-ossification.Bathole ukuthi imodeli yokufunda ejulile ye-CNN ibonise ukusebenza kwesilinganiso esifanayo sobudala njengemodeli evamile yokuhlehla.Ucwaningo lukaGuo et al. [15] luhlole ukusebenza kwezigaba zokubekezelela iminyaka kobuchwepheshe be-CNN ngokusekelwe kuma-orthophotos wamazinyo, futhi imiphumela yemodeli ye-CNN yafakazela ukuthi abantu baphumelele kakhulu ekusebenzeni kwabo ngokwezigaba zeminyaka.
Iningi lezifundo ngokulinganisa iminyaka kusetshenziswa ukufunda komshini zisebenzisa izindlela zokufunda ezijulile13,14,15,16,17,18,19,20.Isilinganiso sobudala esisekelwe ekufundeni okujulile kubikwa ukuthi sinembe kakhulu kunezindlela zendabuko.Kodwa-ke, le ndlela inikeza ithuba elincane lokwethula isisekelo sesayensi sezilinganiso zobudala, njengezinkomba zeminyaka ezisetshenziswa ezilinganisweni.Kuphinde kube nombango wezomthetho mayelana nokuthi ngubani ohlolayo.Ngakho-ke, ukulinganisa ubudala okusekelwe ekufundeni okujulile kunzima ukwamukela iziphathimandla zokuphatha nezokwahlulela.Ukumbiwa kwedatha (DM) kuyindlela engakwazi ukuthola nje kuphela okulindelekile kodwa nolwazi olungalindelekile njengendlela yokuthola ukuhlobana okuwusizo phakathi kwenani elikhulu ledatha6,21,22.Ukufunda ngomshini kuvame ukusetshenziswa ekumbiweni kwedatha, futhi kokubili ukumbiwa kwedatha nokufunda komshini kusebenzisa ama-algorithms angukhiye afanayo ukuze kutholwe amaphethini kudatha.Ukulinganisa kweminyaka kusetshenziswa ukuthuthukiswa kwamazinyo kusekelwe ekuhloleni komhloli ukuvuthwa kwamazinyo okuhlosiwe, futhi lokhu kuhlola kuvezwa njengesiteji sezinyo ngalinye eliqondiwe.I-DM ingasetshenziswa ukuze kuhlaziywe ukuhlobana phakathi kwesigaba sokuhlolwa kwamazinyo kanye neminyaka yangempela futhi inamandla okuthatha indawo yokuhlaziywa kwezibalo okuvamile.Ngakho-ke, uma sisebenzisa amasu e-DM ekulinganisweni kweminyaka, singasebenzisa ukufunda ngomshini esilinganisweni seminyaka yobudala ngaphandle kokukhathazeka ngesibopho sezomthetho.Ucwaningo oluningi lokuqhathanisa luye lwanyatheliswa kwezinye izindlela ezingase zibe khona ezindleleni zendabuko ezisetshenziswa ezenzweni zesayensi nezindlela ezisekelwe ku-EBM zokunquma ubudala bamazinyo.U-Shen et al23 ubonise ukuthi imodeli ye-DM inembe kakhulu kunefomula yeCamerer evamile.U-Galibourg et al24 basebenzise izindlela ezihlukene ze-DM ukuze babikezele ubudala ngokombandela we-Demirdjian25 futhi imiphumela yabonisa ukuthi indlela ye-DM yayisebenza kahle kakhulu kunokusebenzisa izindlela zikaDemirdjian nezaWillems ekulinganiseni iminyaka yabantu baseFrance.
Ukulinganisa iminyaka yamazinyo yentsha yase-Korea kanye nabantu abadala abasebasha, indlela ka-Lee yesi-4 isetshenziswa kakhulu kumkhuba waseKorea.Le ndlela isebenzisa ukuhlaziya kwezibalo okuvamile (okufana nokuhlehla kaningi) ukuze ihlole ubudlelwano phakathi kwezihloko zesi-Korean nobudala bokulandelana kwezikhathi.Kulolu cwaningo, izindlela zokulinganisa ubudala ezitholwe kusetshenziswa izindlela zezibalo zendabuko zichazwa “njengezindlela zendabuko.”Indlela ka-Lee iyindlela evamile, futhi ukunemba kwayo kuqinisekiswe ngu-Oh et al.5;kodwa-ke, ukusebenza kokulinganisa kweminyaka okususelwe kumodeli ye-DM kumkhuba wase-Korean wokuphenya kusangabazeka.Umgomo wethu bekuwukuqinisekisa ngokwesayensi ukuba wusizo okungaba khona kokulinganisa kweminyaka ngokusekelwe kumodeli ye-DM.Injongo yalolu cwaningo bekuwukuthi (1) ukuqhathanisa ukunemba kwamamodeli amabili e-DM ekulinganiseni ubudala bamazinyo kanye (2) nokuqhathanisa ukusebenza kwezigaba kwamamodeli angu-7 DM lapho eneminyaka engu-18 ubudala nalawo atholakala kusetshenziswa izindlela zezibalo zendabuko Ukuvuthwa kwesibili. kanye ne-molars yesithathu kuyo yomibili imihlathi.
Izindlela nokuchezuka okujwayelekile kweminyaka yobudala bokulandelana ngezigaba kanye nohlobo lwamazinyo kuboniswa ku-inthanethi kokuthi Ithebula Le-Supplementary S1 (isethi yokuqeqesha), Ithebula Le-Supplementary S2 (isethi yokuhlola yangaphakathi), kanye neThebula Lesengezo S3 (isethi yokuhlola yangaphandle).Amanani e-kappa okuthembeka kwe-intra- kanye ne-interrobserver atholwe kusethi yokuqeqeshwa abengu-0.951 kanye no-0.947, ngokulandelanayo.Amanani we-P kanye nezikhawu zokuzethemba ezingu-95% zamanani e-kappa aboniswa kuthebula le-inthanethi elingeziwe elingu-S4.Inani le-kappa lihunyushwe ngokuthi “cishe liphelele”, ngokuhambisana nemibandela ye-Landis ne-Koch26.
Uma kuqhathaniswa iphutha eliphelele elisho ukuthini (MAE), indlela yendabuko idlula kancane imodeli ye-DM yabo bonke ubulili nakusethi yokuhlola yangaphandle yabesilisa, ngaphandle kwe-multilayer perceptron (MLP).Umehluko phakathi kwemodeli yendabuko kanye nemodeli ye-DM kusethi yokuhlola yangaphakathi ye-MAE ubuyiminyaka engu-0.12–0.19 kwabesilisa kanye neminyaka engu-0.17–0.21 kwabesifazane.Kubhethri lokuhlola langaphandle, umehluko mncane (iminyaka engu-0.001–0.05 kwabesilisa kanye neminyaka engu-0.05–0.09 kwabesifazane).Ukwengeza, i-root mean square error (RMSE) ingaphansi kancane kunendlela evamile, inomehluko omncane (0.17–0.24, 0.2–0.24 kusethi yokuhlola yangaphakathi yabesilisa, kanye no-0.03–0.07, 0.04–0.08 kusethi yokuhlola yangaphandle).).I-MLP ibonisa ukusebenza okungcono kancane kune-Single Layer Perceptron (SLP), ngaphandle kwasesimeni sesethi yokuhlola yangaphandle yabesifazane.Ku-MAE ne-RMSE, ukuhlolwa kwangaphandle kusetha amaphuzu aphezulu kunokuhlola kwangaphakathi okusethelwe bonke ubulili namamodeli.Wonke ama-MAE kanye ne-RMSE akhonjisiwe kuThebula 1 nakuMfanekiso 1.
I-MAE ne-RMSE yamamodeli endabuko kanye nedatha yokuhlehla kwezimayini.Kusho iphutha eliphelele MAE, impande isho iphutha lesikwele i-RMSE, i-perceptron yesendlalelo esisodwa SLP, i-perceptron ye-multilayer MLP, indlela ye-CM yendabuko.
Ukusebenza kokuhlelwa (okunomkhawulo weminyaka engu-18) wamamodeli wendabuko kanye ne-DM kuye kwaboniswa ngokuya ngokuzwela, ukucaciswa, inani elihle lokubikezela (PPV), inani elibikezela elibi (NPV), kanye nendawo engaphansi kwejika lesici somamukeli (AUROC) 27 (Ithebula 2, Umfanekiso 2 kanye Nomfanekiso Owengeziwe 1 ku-inthanethi).Mayelana nokuzwela kwebhethri lokuhlola langaphakathi, izindlela ezivamile zisebenze kahle kakhulu phakathi kwabesilisa futhi okubi nakakhulu kwabesifazane.Nokho, umehluko ekusebenzeni ngokwezigaba phakathi kwezindlela zendabuko kanye ne-SD ngu-9.7% kwabesilisa (MLP) kanye no-2.4% kuphela kwabesifazane (XGBoost).Phakathi kwamamodeli e-DM, ukwehla kwezinto (LR) kubonise ukuzwela okungcono kubo bobabili ubulili.Mayelana nokucaciswa kwesethi yokuhlola yangaphakathi, kubonwe ukuthi amamodeli amane e-SD asebenze kahle kwabesilisa, kuyilapho imodeli evamile yenza kangcono kwabesifazane.Umehluko ekusebenzeni ngezigaba kwabesilisa nabesifazane ngu-13.3% (MLP) no-13.1% (MLP), ngokulandelana, okubonisa ukuthi umehluko ekusebenzeni ngezigaba phakathi kwamamodeli wedlula ukuzwela.Phakathi kwamamodeli e-DM, umshini wokusekelayo we-vector (SVM), isihlahla sesinqumo (DT), namamodeli wehlathi elingahleliwe (RF) asebenze kahle kakhulu kwabesilisa, kuyilapho imodeli ye-LR yenza kahle kakhulu kwabesifazane.I-AUROC yemodeli yendabuko kanye nawo wonke amamodeli we-SD yayinkulu kuno-0.925 (umakhelwane oseduze kakhulu no-k (KNN) emadodeni), ebonisa ukusebenza okuhle kakhulu ngezigaba ekubandlululeni amasampuli aneminyaka engu-1828.Kusethi yokuhlola yangaphandle, kube nokwehla ekusebenzeni kwezigaba ngokuya kokuzwela, ukucaciswa kanye ne-AUROC uma kuqhathaniswa nesethi yokuhlola yangaphakathi.Ngaphezu kwalokho, umehluko wokuzwela nokucaciswa phakathi kokusebenza kwezigaba kwamamodeli angcono kakhulu namabi kakhulu wawusukela ku-10% ukuya ku-25% futhi wawumkhulu kunomehluko kusethi yokuhlola yangaphakathi.
Ukuzwela kanye nokucaciswa kwamamodeli okuhlukaniswa kwedatha yezimayini uma kuqhathaniswa nezindlela zendabuko ezinokunqanyulwa kweminyaka eyi-18.I-KNN k umakhelwane oseduze, umshini we-vector osekela i-SVM, ukuhlehla kwezinto ze-LR, isihlahla sesinqumo se-DT, ihlathi elingahleliwe le-RF, i-XGB XGBoost, i-MLP ye-multilayer perceptron, indlela ye-CM yendabuko.
Isinyathelo sokuqala kulolu cwaningo sasiwukuqhathanisa ukunemba kwezilinganiso zeminyaka yamazinyo ezitholwe kumamodeli ayisikhombisa e-DM nalawo atholwe kusetshenziswa ukuhlehla okuvamile.I-MAE ne-RMSE zihlolwe ngamasethi okuhlola angaphakathi kubo bobabili ubulili, futhi umehluko phakathi kwendlela evamile kanye nemodeli ye-DM wawusukela ezinsukwini ezingu-44 kuya kwezingu-77 ze-MAE futhi kusukela ezinsukwini ezingu-62 kuya kwezingu-88 ze-RMSE.Nakuba indlela yendabuko yayinembe kakhudlwana kulolu cwaningo, kunzima ukuphetha ngokuthi umehluko omncane kangako unokubaluleka komtholampilo noma okungokoqobo.Le miphumela ibonisa ukuthi ukunemba kwesilinganiso sobudala bamazinyo kusetshenziswa imodeli ye-DM kucishe kufane nendlela evamile.Ukuqhathanisa okuqondile nemiphumela yocwaningo lwangaphambilini kunzima ngoba alukho ucwaningo oluqhathanise ukunemba kwamamodeli e-DM nezindlela zezibalo zendabuko kusetshenziswa indlela efanayo yokuqopha amazinyo ebangeni lobudala elifanayo njengakulolu cwaningo.U-Galibourg et al24 uqhathanise i-MAE ne-RMSE phakathi kwezindlela ezimbili zendabuko (indlela ye-Demirjian25 ne-Willems method29) kanye namamodeli we-DM angu-10 kubantu baseFrance abaneminyaka engu-2 kuya kwengu-24.Babike ukuthi wonke amamodeli e-DM ayenembe kakhulu kunezindlela zendabuko, anomehluko weminyaka engu-0.20 no-0.38 ku-MAE kanye neminyaka engu-0.25 kanye neminyaka engu-0.47 ku-RMSE uma kuqhathaniswa nezindlela ze-Willems ne-Demirdjian, ngokulandelanayo.Umehluko phakathi kwemodeli ye-SD nezindlela zendabuko eziboniswe ocwaningweni lwase-Halibourg ucabangela imibiko eminingi30,31,32,33 yokuthi indlela ye-Demirdjian ayilinganiseli ngokunembile iminyaka yamazinyo kubantu ngaphandle kwabase-French Canadian lapho ucwaningo lwalusekelwe khona.kulolu cwaningo.U-Tai et al 34 wasebenzisa i-algorithm ye-MLP ukubikezela iminyaka yamazinyo kusukela ezithombeni ze-orthodontic zaseShayina ezingu-1636 futhi waqhathanisa ukunemba kwayo nemiphumela yendlela ye-Demirjian ne-Willems.Babike ukuthi i-MLP inokunemba okuphezulu kunezindlela zendabuko.Umehluko phakathi kwendlela ye-Demirdjian kanye nendlela yendabuko ingu-<0.32 iminyaka, futhi indlela ye-Willems iminyaka engu-0.28, efana nemiphumela yocwaningo lwamanje.Imiphumela yalezi zifundo zangaphambilini24,34 ibuye ihambisane nemiphumela yocwaningo lwamanje, futhi ukunemba kokulinganisa kweminyaka yemodeli ye-DM kanye nendlela yendabuko kuyafana.Nokho, ngokusekelwe emiphumeleni ethulwayo, singaphetha ngokuqapha ngokuthi ukusetshenziswa kwamamodeli e-DM ukuze silinganisele ubudala kungase kuthathele indawo izindlela ezikhona ngenxa yokuntuleka kokuqhathanisa nokubhekisela ezifundweni zangaphambilini.Izifundo zokulandelela kusetshenziswa amasampula amakhulu ziyadingeka ukuze kuqinisekiswe imiphumela etholwe kulolu cwaningo.
Phakathi kwezifundo ezihlola ukunemba kwe-SD ekulinganiseni ubudala bamazinyo, ezinye zibonise ukunemba okuphezulu kunocwaningo lwethu.UStepanovsky et al 35 basebenzise amamodeli e-SD angama-22 kuma-radiographs wepanoramic wezakhamuzi zase-Czech ezingu-976 ezineminyaka engu-2.7 kuya kwengu-20.5 futhi bahlola ukunemba kwemodeli ngayinye.Bahlole ukuthuthukiswa kwengqikithi yamazinyo angapheli angu-16 aphezulu naphansi kwesokunxele besebenzisa indlela yokuhlukanisa ehlongozwa nguMoorrees et al 36.I-MAE isukela ku-0.64 kuya ku-0.94 iminyaka kanti i-RMSE isukela ku-0.85 kuya ku-1.27 iminyaka, enembe kakhulu kunezinhlobo ezimbili ze-DM ezisetshenziswe kulolu cwaningo.U-Shen et al23 basebenzise indlela ye-Cameriere ukuze balinganisele ubudala bamazinyo amazinyo ayisikhombisa ahlala njalo ku-mandible kwesokunxele ezakhamuzini zase-China ezisempumalanga ezineminyaka engu-5 kuya kwengu-13 ubudala futhi wayiqhathanisa neminyaka elinganiselwe kusetshenziswa ukuhlehla komugqa, i-SVM ne-RF.Babonise ukuthi womathathu amamodeli e-DM anokunemba okuphezulu uma kuqhathaniswa nefomula evamile ye-Cameriere.I-MAE kanye ne-RMSE ocwaningweni luka-Shen abephansi kunalawo akumodeli ye-DM kulolu cwaningo.Ukunemba okwandisiwe kwezifundo zikaStepanovsky et al.35 kanye noShen et al.23 kungenzeka kungenxa yokufakwa kwezifundo ezincane kumasampula azo ocwaningo.Ngenxa yokuthi izilinganiso zobudala zabahlanganyeli abanamazinyo akhulayo ziba nembe kakhudlwana njengoba inani lamazinyo likhula ngesikhathi sokuthuthukiswa kwamazinyo, ukunemba kwendlela yokulinganisa yobudala ewumphumela kungase kube sengozini lapho ababambiqhaza bocwaningo besebancane.Ukwengeza, iphutha le-MLP ekulinganisweni kweminyaka lincane kancane kune-SLP, okusho ukuthi i-MLP inembe kakhulu kune-SLP.I-MLP ithathwa njengengcono kancane ekulinganisweni kweminyaka, mhlawumbe ngenxa yezendlalelo ezifihliwe ku-MLP38.Nokho, kukhona okuhlukile kusampula yangaphandle yabesifazane (SLP 1.45, MLP 1.49).Ukuthola ukuthi i-MLP inembe kakhulu kune-SLP ekuhloleni ubudala kudinga izifundo ezengeziwe zokubheka emuva.
Ukusebenza ngokwezigaba kwemodeli ye-DM kanye nendlela evamile enkathini yeminyaka eyi-18 nakho kwaqhathaniswa.Wonke amamodeli e-SD ahloliwe nezindlela zendabuko kusethi yokuhlola yangaphakathi zibonise amazinga amukelekayo okucwasa kusampula elineminyaka engu-18 ubudala.Ukuzwela kwabesilisa nabesifazane bekungaphezulu kwama-87.7% nama-94.9%, ngokulandelana, futhi ukucaciswa kwakungaphezulu kwama-89.3% nama-84.7%.I-AUROC yawo wonke amamodeli ahloliwe nayo idlula i-0.925.Ngokolwazi lwethu, alukho ucwaningo oluhlole ukusebenza kwemodeli ye-DM ekuhlukaniseni iminyaka engu-18 ngokusekelwe ekuvuthweni kwamazinyo.Singaqhathanisa imiphumela yalolu cwaningo nokusebenza ngezigaba kwamamodeli okufunda okujulile kuma-panoramic radiographs.U-Guo et al.15 babale ukusebenza kwezigaba kwemodeli yokufunda okujulile esekelwe ku-CNN kanye nendlela eyenziwa ngesandla esekelwe endleleni ka-Demirjian yomkhawulo othile wobudala.Ukuzwela kanye nokucaciswa kwendlela ye-manual kwakungama-87.7% kanye ne-95.5%, ngokulandelana, futhi ukuzwela nokucaciswa kwemodeli ye-CNN kudlule i-89.2% ne-86.6%, ngokulandelana.Baphethe ngokuthi amamodeli okufunda ajulile angangena esikhundleni noma enze ngaphezu kokuhlola okwenziwa ngesandla ekuhlukaniseni izilinganiso zobudala.Imiphumela yalolu cwaningo ibonise ukusebenza kwezigaba okufanayo;Kukholakala ukuthi ukuhlukaniswa kusetshenziswa amamodeli e-DM kungangena esikhundleni sezindlela zezibalo zendabuko zokulinganisa ubudala.Phakathi kwamamodeli, i-DM LR yayiyimodeli engcono kakhulu mayelana nokuzwela kwesampula yabesilisa nokuzwela kanye nokucaciswa kwesampula yabesifazane.I-LR ikleliswe endaweni yesibili ngokuqondene nabesilisa.Ngaphezu kwalokho, i-LR ibhekwa njengenye yamamodeli e-DM35 asebenziseka kalula futhi ayiyinkimbinkimbi futhi enzima ukuyicubungula.Ngokusekelwe kule miphumela, i-LR ithathwe njengemodeli engcono kakhulu yokuhlukanisa izingane ezineminyaka engu-18 kubantu base-Korea.
Sekukonke, ukunemba kwesilinganiso sobudala noma ukusebenza kwezigaba kusethi yokuhlola yangaphandle bekukubi noma kuphansi uma kuqhathaniswa nemiphumela ekusethi yokuhlola yangaphakathi.Eminye imibiko ibonisa ukuthi ukunemba kwezigaba noma ukusebenza kahle kuyehla lapho izilinganiso zobudala ezisuselwe kusibalo sabantu base-Korea zisetshenziswa kubantu base-Japan5,39, futhi iphethini efanayo itholwe ocwaningweni lwamanje.Lokhu kuthambekela kokuwohloka kwabonwa nakumodeli ka-DM.Ngakho-ke, ukulinganisa ngokunembile iminyaka yobudala, ngisho nalapho kusetshenziswa i-DM enqubweni yokuhlaziya, izindlela ezithathwe kudatha yabantu bomdabu, njengezindlela ezivamile, kufanele zikhethwe5,39,40,41,42.Njengoba kungacaci ukuthi amamodeli okufunda ajulile angabonisa yini izitayela ezifanayo, izifundo eziqhathanisa ukunemba kwezigaba nokusebenza kahle kusetshenziswa izindlela ezivamile, amamodeli e-DM, namamodeli okufunda ajulile kumasampula afanayo ayadingeka ukuze kuqinisekiswe ukuthi ubuhlakani bokwenziwa bungakwazi yini ukunqoba lokhu kuhlukana kwezinhlanga eminyakeni elinganiselwe.ukuhlola.
Sibonisa ukuthi izindlela zendabuko zingathathelwa indawo yisilinganiso sobudala ngokusekelwe kumodeli ye-DM kumkhuba wokuhlola iminyaka yobudala e-Korea.Siphinde sathola ithuba lokuqalisa ukufunda ngomshini ukuze kuhlolwe iminyaka yobudala.Kodwa-ke, kunemikhawulo ecacile, njengenani elinganele labahlanganyeli kulolu cwaningo ukuze linqume ngokucacile imiphumela, kanye nokuntuleka kwezifundo zangaphambilini zokuqhathanisa nokuqinisekisa imiphumela yalolu cwaningo.Ngokuzayo, izifundo ze-DM kufanele zenziwe ngezinombolo ezinkulu zamasampula kanye nenani labantu abahlukahlukene ukuze kuthuthukiswe ukusebenza kwayo okungokoqobo uma kuqhathaniswa nezindlela zendabuko.Ukuze kuqinisekiswe ukuthi kungenzeka yini ukusebenzisa ubuhlakani bokwenziwa ukulinganisa iminyaka kumanani abantu abaningi, ucwaningo lwesikhathi esizayo luyadingeka ukuze kuqhathaniswe ukunemba kwezigaba nokusebenza kahle kwe-DM kanye namamodeli okufunda ajulile nezindlela zendabuko kumasampuli afanayo.
Ucwaningo lusebenzise izithombe ze-orthographic ezingu-2,657 eziqoqwe kubantu abadala baseKorea nabaseJapane abaneminyaka engu-15 kuya kwengama-23.Ama-radiographs aseKorea ahlukaniswe amasethi okuqeqesha angu-900 (iminyaka engu-19.42 ± 2.65) kanye namasethi okuhlola angaphakathi angu-900 (iminyaka engu-19.52 ± 2.59).Isethi yokuqeqeshwa yaqoqwa esikhungweni esisodwa (Isibhedlela iSeoul St. Mary's), futhi isethi yokuhlola yaqoqwa ezikhungweni ezimbili (i-Seoul National University Dental Hospital kanye ne-Yonsei University Dental Hospital).Siphinde saqoqa ama-radiographs angu-857 kwenye idatha yabantu (i-Iwate Medical University, Japan) ukuze ihlolwe ngaphandle.Ama-radiographs ezifundo zesi-Japanese (iminyaka engu-19.31 ± 2.60) akhethwe njengesethi yokuhlola yangaphandle.Idatha iqoqwe ngokubuyela emuva ukuze kuhlaziywe izigaba zokuthuthukiswa kwamazinyo kuma-panoramic radiographs athathwe ngesikhathi sokwelashwa kwamazinyo.Yonke idatha eqoqwe yayingaziwa ngaphandle kobulili, usuku lokuzalwa nosuku lwe-radiograph.Imibandela yokufakwa nokukhipha inyumbazane yayifana nezifundo ezishicilelwe ngaphambili 4 , 5 .Iminyaka yangempela yesampula ibalwa ngokukhipha idethi yokuzalwa kusukela ngosuku okwathathwa ngalo i-radiograph.Iqembu lesampula lahlukaniswa laba amaqembu obudala obuyisishiyagalolunye.Ukusatshalaliswa kweminyaka yobudala nobulili kuboniswa kuThebula 3 Lolu cwaningo lwenziwe ngokuhambisana neSimemezelo sase-Helsinki futhi sagunyazwa Ibhodi Lokubuyekeza Isikhungo (IRB) sase-Seoul St. Mary's Hospital ye-Catholic University of Korea (KC22WISI0328).Ngenxa yedizayini ye-retrospective yalolu cwaningo, imvume enolwazi ayikwazanga ukutholwa kuzo zonke iziguli ezihlolwa nge-radiographic ngezinjongo zokwelapha.Isibhedlela i-Seoul Korea University St. Mary's Hospital (IRB) siyekele imfuneko yemvume unolwazi.
Izigaba zokuthuthuka ze-bimaxillary molars yesibili neyesithathu zahlolwa ngokwemibandela ye-Demircan25.Kwakhethwa izinyo elilodwa kuphela uma kutholakala uhlobo olufanayo lwezinyo ohlangothini lwesobunxele nolwesokudla lomhlathi ngamunye.Uma amazinyo angama-homologous nhlangothi zombili ebesezigabeni zokukhula ezihlukene, izinyo elinesigaba sokukhula esiphansi lalikhethwa ukuze kuphendule ukungaqiniseki eminyakeni elinganiselwe.Ama-radiographs ayikhulu akhethwe ngokungahleliwe kusukela kusethi yokuqeqeshwa atholwe izibukeli ezimbili ezinolwazi ukuze kuhlolwe ukwethembeka kwe-interobserver ngemva kokulinganisa kusengaphambili ukuze kunqunywe isigaba sokuvuthwa kwamazinyo.Ukuthembeka kwe-Intraobserver kwahlolwa kabili ngezikhathi zezinyanga ezintathu yisibukeli esiyinhloko.
Isigaba sobulili nesokuthuthuka se-molars yesibili neyesithathu yomhlathi ngamunye kusethi yokuqeqeshwa silinganiselwa ngumuntu obukele oyinhloko oqeqeshwe ngamamodeli ahlukene we-DM, futhi iminyaka yangempela yabekwa njengenani eliqondiwe.Amamodeli e-SLP nawe-MLP, asetshenziswa kakhulu ekufundeni komshini, ahlolwe ngokumelene nama-algorithms wokuhlehla.Imodeli ye-DM ihlanganisa imisebenzi elandelanayo isebenzisa izigaba zokukhula zamazinyo amane futhi ihlanganisa le datha ukuze ilinganise iminyaka.I-SLP iyinethiwekhi ye-neural elula kakhulu futhi ayinazo izendlalelo ezifihliwe.I-SLP isebenza ngokususelwe ekudlulisweni kwe-threshold phakathi kwamanodi.Imodeli ye-SLP ekuhlehleni iyafana ngokwezibalo nokuhlehla kwemigqa eminingi.Ngokungafani nemodeli ye-SLP, imodeli ye-MLP inezendlalelo eziningi ezifihliwe ezinemisebenzi yokuvula engaqondile.Ukuhlola kwethu kusebenzise isendlalelo esifihliwe esinama-node afihliwe angama-20 anemisebenzi yokuvula engaqondile.Sebenzisa ukwehla kwe-gradient njengendlela yokuthuthukisa kanye ne-MAE ne-RMSE njengomsebenzi wokulahlekelwa ukuze uqeqeshe imodeli yethu yokufunda yomshini.Imodeli yokuhlehla etholwe kahle kakhulu isetshenziswe kumasethi okuhlola angaphakathi nangaphandle futhi iminyaka yamazinyo yalinganiselwa.
I-algorithm yokuhlukanisa yasungulwa esebenzisa ukuvuthwa kwamazinyo amane kusethi yokuqeqeshwa ukuze kubikezelwe ukuthi isampula lineminyaka engu-18 ubudala noma cha.Ukuze sakhe imodeli, sithole ama-algorithms okufunda komshini omelelayo ayisikhombisa6,43: (1) LR, (2) KNN, (3) SVM, (4) DT, (5) RF, (6) XGBoost, kanye (7) MLP. .I-LR ingenye yama-algorithms okuhlukanisa asetshenziswa kakhulu44.Kuyi-algorithm yokufunda egadiwe esebenzisa ukuhlehla ukuze ibikezele amathuba edatha yesigaba esithile ukusuka ku-0 kuye ku-1 futhi ihlukanise idatha njengengxenye yesigaba okungenzeka kakhulu ngokususelwe kula mathuba;ikakhulukazi esetshenziselwa ukuhlukanisa kanambambili.I-KNN ingenye yezindlela ezilula zokufunda umshini45.Uma inikezwa idatha yokufaka entsha, ithola idatha ethi k eduze nesethi ekhona bese iwahlukanisa ekilasini ngefrikhwensi ephezulu kakhulu.Sibeka u-3 ngenani lomakhelwane abacatshangelwayo (k).I-SVM i-algorithm eyandisa ibanga phakathi kwamakilasi amabili ngokusebenzisa umsebenzi we-kernel ukuze kwandiswe isikhala esiwumugqa sibe isikhala esingalayini esibizwa ngokuthi izinkambu46.Kule modeli, sisebenzisa i-bias = 1, amandla = 1, ne-gamma = 1 njengama-hyperparameter e-polynomial kernel.I-DT isetshenziswe emikhakheni ehlukahlukene njenge-algorithm yokwehlukanisa idatha yonke esethwe emaqenjini amancane amaningana ngokumelela imithetho yesinqumo esakhiweni sesihlahla47.Imodeli icushwe ngenani elincane lamarekhodi inodi ngayinye engu-2 futhi isebenzisa inkomba ye-Gini njengesilinganiso sekhwalithi.I-RF iyindlela yokuhlanganisa ehlanganisa ama-DT amaningi ukuze kuthuthukiswe ukusebenza kusetshenziswa indlela yokuhlanganisa ye-bootstrap ekhiqiza isihlukanisi esibuthakathaka sesampula ngayinye ngokudweba ngokungahleliwe amasampula anosayizi ofanayo izikhathi eziningi kudathasethi48.Sisebenzise izihlahla eziyi-100, ukujula kwezihlahla eziyi-10, usayizi wenodi obuncane obu-1, kanye nenkomba ye-Gini ye-admixture njengemibandela yokuhlukanisa amanodi.Ukuhlukaniswa kwedatha entsha kunqunywa ivoti eliningi.I-XGBoost i-algorithm ehlanganisa amasu okuthuthukisa isebenzisa indlela ethatha njengedatha yokuqeqesha iphutha phakathi kwamanani angempela nabikezelwe emodeli yangaphambili futhi ikhulise iphutha kusetshenziswa ama-gradient49.Kuyi-algorithm esetshenziswa kabanzi ngenxa yokusebenza kwayo okuhle nokusebenza kahle kwezinsiza, kanye nokuthembeka okuphezulu njengomsebenzi wokulungisa owedlulele.Imodeli ifakwe amasondo angama-400 asekelayo.I-MLP inethiwekhi ye-neural lapho i-perceptron eyodwa noma ngaphezulu yakha khona izendlalelo eziningi ezinengqimba efihliwe eyodwa noma ngaphezulu phakathi kokufakwayo nokukhiphayo38.Ngokusebenzisa lokhu, ungenza ukuhlukanisa okungekona komugqa lapho uma ungeza isendlalelo sokufaka futhi uthole inani lomphumela, inani lomphumela elibikezelwe liqhathaniswa nenani langempela lomphumela futhi iphutha libuye lisakazwe.Sakhe isendlalelo esifihliwe esinama-neurons afihliwe angama-20 kusendlalelo ngasinye.Imodeli ngayinye esiyithuthukisile yasetshenziswa kumasethi angaphakathi nangaphandle ukuze kuhlolwe ukusebenza kwezigaba ngokubala ukuzwela, ukucaciswa, i-PPV, i-NPV, ne-AUROC.Ukuzwela kuchazwa njengesilinganiso sesampula esilinganiselwa eminyakeni engu-18 ubudala noma ngaphezulu kusampula elilinganiselwa elineminyaka engu-18 noma ngaphezulu.Ukucaciswa ingxenye yamasampuli angaphansi kweminyaka engu-18 ubudala kanye nalawo alinganiselwa ukuthi angaphansi kweminyaka engu-18 ubudala.
Izigaba zamazinyo ezihlolwe kusethi yokuqeqeshwa zaguqulwa zaba izigaba zezinombolo zokuhlaziywa kwezibalo.Ukuhlehla kwemigqa eminingi nokuhlelekayo kwenziwa ukuze kuthuthukiswe amamodeli aqagelayo ocansi ngalunye futhi kutholwe amafomula okuhlehla angasetshenziswa ukulinganisa iminyaka yobudala.Sisebenzise lawa mafomula ukulinganisa iminyaka yamazinyo kuwo womabili amasethi okuhlola angaphakathi nangaphandle.Ithebula lesi-4 likhombisa izibonelo zokuhlehla nokuhlela ezisetshenziswe kulolu cwaningo.
Ukuthembeka kwe-Intra- ne-interrobserver kubalwe kusetshenziswa izibalo ze-kappa zika-Cohen.Ukuze sihlole ukunemba kwe-DM namamodeli angokwesiko okuhlehla, sibale i-MAE ne-RMSE sisebenzisa iminyaka elinganiselwe neyangempela yamasethi okuhlola angaphakathi nangaphandle.Lawa maphutha ajwayele ukusetshenziselwa ukuhlola ukunemba kwezibikezelo zemodeli.Uma iphutha liba lincane, kuphakama ukunemba kwesibikezelo24.Qhathanisa i-MAE ne-RMSE yamasethi okuhlola angaphakathi nangaphandle abalwa kusetshenziswa i-DM nokuhlehla okuvamile.Ukusebenza kokuhlelwa kokunqanyulwa kweminyaka engu-18 kuzibalo ezivamile kwahlolwa kusetshenziswa ithebula lezimo eziphuthumayo ezingu-2 × 2.Ukuzwela okubaliwe, ukucaciswa, i-PPV, i-NPV, ne-AUROC yesethi yokuhlola kuqhathaniswe namanani akalwayo emodeli yokuhlukaniswa kwe-DM.Idatha ichazwa njengokushiwo ± ukuchezuka okujwayelekile noma inombolo (%) kuye ngezici zedatha.Amanani we-P anezinhlangothi ezimbili <0.05 athathwe njengebalulekile ngokwezibalo.Konke ukuhlaziya kwezibalo okuvamile kwenziwa kusetshenziswa inguqulo ye-SAS 9.4 (SAS Institute, Cary, NC).Imodeli yokuhlehla ye-DM yasetshenziswa ku-Python kusetshenziswa i-Keras50 2.2.4 backend kanye ne-Tensorflow51 1.8.0 ngokukhethekile ekusebenzeni kwezibalo.Imodeli yokuhlukaniswa kwe-DM yasetshenziswa ku-Waikato Knowledge Analysis Environment kanye ne-Konstanz Information Miner (KNIME) 4.6.152 platform yokuhlaziya.
Ababhali bayavuma ukuthi idatha esekela iziphetho zocwaningo ingatholakala ku-athikili kanye nezinto ezengeziwe.Amasethi edatha akhiqizwe kanye/noma ahlaziywa phakathi nocwaningo ayatholakala kumbhali ohambisanayo ngesicelo esifanele.
Ritz-Timme, S. et al.Ukuhlolwa kweminyaka: isimo sobuciko ukuze kuhlangatshezwane nezidingo ezithile zokwenziwa kophenyo.amazwe ngamazwe.J. Umuthi wezomthetho.113, 129–136 (2000).
Schmeling, A., Reisinger, W., Geserik, G., kanye no-Olze, A. Isimo samanje sokuhlolwa kweminyaka yobudala yezifundo eziphilayo ngezinjongo zokushushiswa kobugebengu.I-Forensics.umuthi.I-Pathology.1, 239–246 (2005).
Pan, J. et al.Indlela elungisiwe yokuhlola iminyaka yamazinyo yezingane ezineminyaka engu-5 kuya kwengu-16 empumalanga yeShayina.emtholampilo.Inhlolovo yomlomo.25, 3463–3474 (2021).
U-Lee, u-SS njll. Ukulandelana kwezikhathi kokuthuthukiswa kwe-molars yesibili neyesithathu kuma-Koreans kanye nokusetshenziswa kwayo kokuhlolwa kweminyaka yobudala.amazwe ngamazwe.J. Umuthi wezomthetho.124, 659–665 (2010).
Oh, S., Kumagai, A., Kim, SY kanye no-Lee, SS Ukunemba kwesilinganiso sobudala kanye nokulinganisa komkhawulo weminyaka engu-18 ngokusekelwe ekuvuthweni kwe-molars yesibili neyesithathu kuma-Korean naseJapane.I-PLoS ONE 17, e0271247 (2022).
UKim, JY, et al.Ukuhlaziywa kwedatha okusekelwe ekufundeni komshini ngaphambi kokusebenza kungabikezela umphumela wokwelashwa kokuhlinzwa kokulala ezigulini ezine-OSA.isayensi.Bika 11, 14911 (2021).
U-Han, M. et al.Isilinganiso sobudala esinembile esivela ekufundeni ngomshini noma ngaphandle kokungenelela komuntu?amazwe ngamazwe.J. Umuthi wezomthetho.136, 821–831 (2022).
U-Khan, S. kanye no-Shaheen, M. Ukusuka Emayini Yedatha kuya Emayini Yedatha.J. Ulwazi.isayensi.https://doi.org/10.1177/01655515211030872 (2021).
Khan, S. and Shaheen, M. WisRule: The First Cognitive Algorithm for Association Rule Mining.J. Ulwazi.isayensi.https://doi.org/10.1177/01655515221108695 (2022).
U-Shaheen M. no-Abdullah U. Karm: Ukumbiwa kwedatha evamile okusekelwe emithethweni yokuhlanganisa esekelwe komongo.bala.Math.qhubeka.68, 3305–3322 (2021).
U-Muhammad M., u-Rehman Z., u-Shaheen M., u-Khan M. no-Habib M. Ukutholwa kokufana kwe-semantic okusekelwe ekufundeni okujulile kusetshenziswa idatha yombhalo.yazisa.ubuchwepheshe.ukulawula.https://doi.org/10.5755/j01.itc.49.4.27118 (2020).
Tabish, M., Tanoli, Z., kanye no-Shahin, M. A uhlelo lokubona umsebenzi kumavidiyo ezemidlalo.i-multimedia.Izinhlelo zokusebenza zamathuluzi https://doi.org/10.1007/s11042-021-10519-6 (2021).
Halabi, SS et al.I-RSNA Machine Learning Challenge in Pediatric Bone Age.I-Radiology 290, 498–503 (2019).
Li, Y. et al.Isilinganiso seminyaka ye-forensic kusukela kuma-X-reyi e-pelvic kusetshenziswa ukufunda okujulile.I-EURO.imisebe.29, 2322–2329 (2019).
Guo, YC, et al.Ukuhlukaniswa kweminyaka okunembile kusetshenziswa izindlela zemanuwali kanye namanethiwekhi ajulile e-convolutional neural avela ezithombeni ze-orthographic projection.amazwe ngamazwe.J. Umuthi wezomthetho.135, 1589–1597 (2021).
Alabama Dalora et al.Isilinganiso sobudala bethambo kusetshenziswa izindlela ezihlukile zokufunda zomshini: ukubuyekezwa kwezincwadi okuhlelekile nokuhlaziywa kwe-meta.I-PLoS ONE 14, e0220242 (2019).
U-Du, H., Li, G., Cheng, K., kanye no-Yang, J. Isilinganiso sobudala esicaciswe ngabantu base-Afrika baseMelika namaShayina esisekelwe kumthamo wegumbi le-pulp le-molars yokuqala kusetshenziswa i-cone-beam computed tomography.amazwe ngamazwe.J. Umuthi wezomthetho.136, 811–819 (2022).
U-Kim S., u-Lee YH, u-Noh YK, u-Park FK kanye no-Oh KS Ukunquma amaqembu eminyaka yobudala babantu abaphilayo kusetshenziswa izithombe zokwenziwa ezisekelwe kubuhlakani bokwenziwa zemolars yokuqala.isayensi.Bika 11, 1073 (2021).
I-Stern, D., Payer, C., Giuliani, N., kanye no-Urschler, M. Ukulinganisa kweminyaka okuzenzakalelayo kanye nokuhlukaniswa kweminyaka eminingi kusukela kudatha ye-MRI ehlukahlukene.IEEE J. Biomed.Izaziso Zezempilo.23, 1392–1403 (2019).
U-Cheng, Q., Ge, Z., Du, H. kanye no-Li, G. Ukulinganisa kweminyaka okusekelwe ekuhlukaniseni kwegumbi le-3D pulp ye-molars yokuqala kusuka ku-cone beam computed tomography ngokuhlanganisa ukufunda okujulile namasethi amaleveli.amazwe ngamazwe.J. Umuthi wezomthetho.135, 365–373 (2021).
Wu, WT, et al.Ukumbiwa kwedatha kudatha enkulu yomtholampilo: isizindalwazi esivamile, izinyathelo, namamodeli wezindlela.Umhlaba.umuthi.insiza.8, 44 (2021).
Yang, J. et al.Isingeniso Semininingo Egciniwe Yezokwelapha kanye Nobuchwepheshe Bezimayini Zedatha Esikhathini Esikhulu Sedatha.J. Avid.Umuthi oyisisekelo.13, 57–69 (2020).
Shen, S. et al.Indlela ye-Camerer yokulinganisa iminyaka yamazinyo kusetshenziswa ukufunda ngomshini.I-BMC Oral Health 21, 641 (2021).
Galliburg A. et al.Ukuqhathaniswa kwezindlela ezihlukene zokufunda zomshini zokubikezela ubudala bamazinyo kusetshenziswa indlela yesiteji ye-Demirdjian.amazwe ngamazwe.J. Umuthi wezomthetho.135, 665–675 (2021).
U-Demirdjian, A., Goldstein, H. no-Tanner, JM Uhlelo olusha lokuhlola ubudala bamazinyo.ukuhonqa.isayensi yezinto eziphilayo.45, 211–227 (1973).
U-Landis, JR, no-Koch, GG Izinyathelo zesivumelwano sombukeli kudatha yesigaba.I-Biometrics 33, 159–174 (1977).
Bhattacharjee S, Prakash D, Kim C, Kim HK and Choi HK.Ukuhlaziywa kombhalo, i-morphological and statistical imaging ye-two-dimensional magnetic resonance imaging kusetshenziswa amasu obuhlakani bokwenziwa ukuze kuhlukaniswe izimila zobuchopho eziyinhloko.Ulwazi lwezempilo.insiza.https://doi.org/10.4258/hir.2022.28.1.46 (2022).
Isikhathi sokuthumela: Jan-04-2024