Zikomo pochezera Nature.com. Mtundu wa msakatuli womwe mukugwiritsa ntchito uli ndi chithandizo chochepa cha CSS. Kuti mupeze zotsatira zabwino, tikukulimbikitsani kugwiritsa ntchito mtundu watsopano wa msakatuli wanu (kapena kuzimitsa mawonekedwe ogwirizana mu Internet Explorer). Pakadali pano, kuti tiwonetsetse kuti chithandizo chikupitilira, tikuonetsa tsamba lino popanda kukongoletsa kapena JavaScript.
Mano amaonedwa kuti ndi chizindikiro cholondola kwambiri cha zaka za thupi la munthu ndipo nthawi zambiri amagwiritsidwa ntchito poyesa zaka za forensic. Cholinga chathu chinali kutsimikizira kuyerekezera kwa zaka za mano pogwiritsa ntchito deta poyerekezera kulondola kwa kuyerekezera ndi magwiridwe antchito a zaka 18 ndi njira zachikhalidwe komanso kuyerekezera kwa zaka pogwiritsa ntchito deta. Ma radiograph okwana 2657 a panoramic anasonkhanitsidwa kuchokera kwa nzika zaku Korea ndi Japan zazaka zapakati pa 15 mpaka 23. Anagawidwa m'magulu ophunzitsira, iliyonse ili ndi ma radiograph 900 aku Korea, ndi seti yoyesera yamkati yokhala ndi ma radiograph 857 aku Japan. Tinayerekeza kulondola kwa kugawa ndi kugwira ntchito bwino kwa njira zachikhalidwe ndi magulu oyesera a mitundu yosungira deta. Kulondola kwa njira yachikhalidwe pa seti yoyesera yamkati ndikokwera pang'ono kuposa kwa chitsanzo chosungira deta, ndipo kusiyana kwake ndi kochepa (chiwerengero chapakati
Kuphunzira kwa makina ndi mtundu wa luntha lochita kupanga (AI) lomwe limaphunzira mobwerezabwereza ndikugawa deta yambiri, kuthetsa mavuto palokha, ndikuyendetsa mapulogalamu a deta. Kuphunzira kwa makina kumatha kupeza njira zothandiza zobisika m'ma data ambiri6. Mosiyana ndi zimenezi, njira zakale, zomwe zimafuna ntchito yambiri komanso nthawi yambiri, zitha kukhala ndi zoletsa pochita ndi deta yambiri yovuta yomwe ndi yovuta kuigwiritsa ntchito pamanja7. Chifukwa chake, maphunziro ambiri achitika posachedwapa pogwiritsa ntchito ukadaulo waposachedwa wa makompyuta kuti achepetse zolakwika za anthu ndikugwiritsa ntchito bwino deta yamitundu yambiri8,9,10,11,12. Makamaka, kuphunzira mozama kwagwiritsidwa ntchito kwambiri pakusanthula zithunzi zachipatala, ndipo njira zosiyanasiyana zowerengera zaka posanthula ma radiograph okha zanenedwa kuti zikukweza kulondola ndi magwiridwe antchito a kuwerengera zaka13,14,15,16,17,18,19,20. Mwachitsanzo, Halabi et al 13 adapanga njira yophunzirira makina yochokera pa ma netiweki a mitsempha ya convolutional (CNN) kuti ayese zaka za mafupa pogwiritsa ntchito ma radiograph a manja a ana. Kafukufukuyu akupereka chitsanzo chomwe chimagwiritsa ntchito kuphunzira kwa makina pazithunzi zachipatala ndipo chikuwonetsa kuti njirazi zitha kupititsa patsogolo kulondola kwa matenda. Li et al14 adayerekeza zaka kuchokera ku zithunzi za X-ray ya m'chiuno pogwiritsa ntchito CNN yophunzirira kwambiri ndipo adaziyerekeza ndi zotsatira za regression pogwiritsa ntchito ossification stage estimate. Adapeza kuti chitsanzo cha deep learning CNN chinawonetsa magwiridwe antchito ofanana ndi a regression model yachikhalidwe. Kafukufuku wa Guo et al. [15] adawunika magwiridwe antchito a CNN pakulekerera zaka kutengera ma orthophotos a mano, ndipo zotsatira za chitsanzo cha CNN zidatsimikizira kuti anthu adachita bwino kuposa momwe amachitira magulu azaka.
Maphunziro ambiri okhudza kuyerekezera zaka pogwiritsa ntchito makina ophunzirira amagwiritsa ntchito njira zophunzirira mozama13,14,15,16,17,18,19,20. Kuyerekeza zaka pogwiritsa ntchito makina ophunzirira mozama akuti ndi kolondola kwambiri kuposa njira zachikhalidwe. Komabe, njira iyi imapereka mwayi wochepa wopereka maziko asayansi a kuyerekezera zaka, monga zizindikiro za zaka zomwe zimagwiritsidwa ntchito mu kuyerekezera. Palinso mkangano wazamalamulo pa yemwe amachita kafukufuku. Chifukwa chake, kuyerekezera zaka pogwiritsa ntchito makina ophunzirira mozama n'kovuta kuvomereza ndi akuluakulu aboma ndi oweruza. Kusanthula deta (DM) ndi njira yomwe ingapeze osati chidziwitso choyembekezeredwa chokha komanso chosayembekezereka ngati njira yopezera mgwirizano wothandiza pakati pa deta yambiri6,21,22. Kuphunzira kwa makina nthawi zambiri kumagwiritsidwa ntchito mu kusanthula deta, ndipo kusanthula deta ndi kuphunzira kwa makina onse amagwiritsa ntchito ma algorithms ofunikira omwewo kuti apeze mawonekedwe mu deta. Kuwerengera zaka pogwiritsa ntchito chitukuko cha mano kumadalira kuwunika kwa woyesa kukula kwa mano owunikira, ndipo kuwunikaku kumawonetsedwa ngati gawo la dzino lililonse lowunikira. Kusanthula deta kungagwiritsidwe ntchito kusanthula mgwirizano pakati pa gawo lowunikira mano ndi zaka zenizeni ndipo kuli ndi kuthekera kosintha kusanthula kwachikhalidwe kwa ziwerengero. Chifukwa chake, ngati tigwiritsa ntchito njira za DM pakuyerekeza zaka, titha kugwiritsa ntchito makina ophunzirira pakuyerekeza zaka popanda kuda nkhawa ndi udindo walamulo. Kafukufuku woyerekeza angapo afalitsidwa pa njira zina zomwe zingatheke m'malo mwa njira zachikhalidwe zomwe zimagwiritsidwa ntchito pamanja zomwe zimagwiritsidwa ntchito pakuchita zamalamulo ndi njira zochokera ku EBM podziwa zaka za mano. Shen et al23 adawonetsa kuti chitsanzo cha DM ndi cholondola kwambiri kuposa njira yachikhalidwe ya Camerer. Galibourg et al24 adagwiritsa ntchito njira zosiyanasiyana za DM kuti adziwiretu zaka malinga ndi muyezo wa Demirdjian25 ndipo zotsatira zake zidawonetsa kuti njira ya DM idapambana njira za Demirdjian ndi Willems pakuyerekeza zaka za anthu aku France.
Pofuna kuwerengera zaka za mano a achinyamata aku Korea ndi achinyamata, njira ya Lee nambala 4 imagwiritsidwa ntchito kwambiri mu kafukufuku wa zamankhwala aku Korea. Njirayi imagwiritsa ntchito kusanthula kwachikhalidwe (monga kubwerezabwereza kwa ziwerengero) kuti iwunike ubale pakati pa anthu aku Korea ndi zaka za nthawi. Mu kafukufukuyu, njira zowerengera zaka zomwe zimapezeka pogwiritsa ntchito njira zachikhalidwe zowerengera zimatanthauzidwa ngati "njira zachikhalidwe." Njira ya Lee ndi njira yachikhalidwe, ndipo kulondola kwake kwatsimikiziridwa ndi Oh et al. 5; komabe, kugwiritsa ntchito kuwerengera zaka kutengera chitsanzo cha DM mu kafukufuku wa zamankhwala aku Korea kukadali kokayikitsa. Cholinga chathu chinali kutsimikizira mwasayansi phindu lomwe lingatheke la kuwerengera zaka kutengera chitsanzo cha DM. Cholinga cha kafukufukuyu chinali (1) kuyerekeza kulondola kwa mitundu iwiri ya DM pakuyerekeza zaka za mano ndi (2) kuyerekeza magwiridwe antchito a mitundu 7 ya DM ali ndi zaka 18 ndi yomwe idapezeka pogwiritsa ntchito njira zachikhalidwe zowerengera Kukula kwa molars yachiwiri ndi yachitatu m'nsagwada zonse ziwiri.
Njira ndi kusiyana kwa msinkhu wa nthawi malinga ndi gawo ndi mtundu wa dzino zikuwonetsedwa pa intaneti mu Supplementary Table S1 (seti yophunzitsira), Supplementary Table S2 (seti yoyesera yamkati), ndi Supplementary Table S3 (seti yoyesera yakunja). Ma values a kappa odalirika mkati ndi mkati mwa owonera omwe adapezeka kuchokera ku seti yophunzitsira anali 0.951 ndi 0.947, motsatana. Ma values a P ndi 95% zodalirika za ma values a kappa akuwonetsedwa mu tebulo lowonjezera la pa intaneti S4. Mtengo wa kappa unatanthauziridwa ngati "wangwiro kwambiri", mogwirizana ndi zofunikira za Landis ndi Koch26.
Poyerekeza cholakwika chapakati (MAE), njira yachikhalidwe imaposa pang'ono chitsanzo cha DM cha amuna ndi akazi onse komanso mu seti yoyesera yakunja ya amuna, kupatula multilayer perceptron (MLP). Kusiyana pakati pa chitsanzo chachikhalidwe ndi chitsanzo cha DM pa seti yoyesera yamkati ya MAE kunali zaka 0.12–0.19 kwa amuna ndi zaka 0.17–0.21 kwa akazi. Pa batire yoyesera yakunja, kusiyanako kumakhala kochepa (zaka 0.001–0.05 kwa amuna ndi zaka 0.05–0.09 kwa akazi). Kuphatikiza apo, cholakwika chapakati chapakati (RMSE) ndi chotsika pang'ono kuposa njira yachikhalidwe, ndi kusiyana kochepa (0.17–0.24, 0.2–0.24 kwa seti yoyesera yamkati ya amuna, ndi 0.03–0.07, 0.04–0.08 kwa seti yoyesera yakunja). ). MLP ikuwonetsa magwiridwe antchito abwinoko pang'ono kuposa Single Layer Perceptron (SLP), kupatula pa seti yoyesera yakunja ya akazi. Pa MAE ndi RMSE, mayeso akunja amapambana kwambiri kuposa mayeso amkati a amuna ndi akazi onse ndi mitundu yonse. MAE ndi RMSE zonse zawonetsedwa mu Table 1 ndi Chithunzi 1.
MAE ndi RMSE ya mitundu yachikhalidwe ndi yosinthira deta. Chilakolako chachikulu cha MAE, cholakwika chachikulu cha root mean square RMSE, single layer perceptron SLP, multilayer perceptron MLP, njira yachikhalidwe ya CM.
Kuchita bwino kwa magulu (ndi kudulidwa kwa zaka 18) kwa mitundu yachikhalidwe ndi ya DM kunawonetsedwa pankhani ya kukhudzidwa, kudziwika, kutsimikizika kwabwino (PPV), kuneneratu koyipa (NPV), ndi dera lomwe lili pansi pa curve yodziwika bwino yolandira (AUROC) 27 (Table 2, Chithunzi 2 ndi Chithunzi Chowonjezera 1 pa intaneti). Ponena za kukhudzidwa kwa batri yoyesera yamkati, njira zachikhalidwe zidagwira ntchito bwino pakati pa amuna komanso zoyipa pakati pa akazi. Komabe, kusiyana kwa magwiridwe antchito pakati pa njira zachikhalidwe ndi SD ndi 9.7% kwa amuna (MLP) ndi 2.4% yokha kwa akazi (XGBoost). Pakati pa mitundu ya DM, logistic regression (LR) idawonetsa kukhudzidwa bwino pakati pa amuna ndi akazi. Ponena za kutsimikizika kwa gulu la mayeso amkati, zidawonedwa kuti mitundu inayi ya SD idachita bwino mwa amuna, pomwe mtundu wachikhalidwe udachita bwino mwa akazi. Kusiyana kwa magwiridwe antchito pakati pa amuna ndi akazi ndi 13.3% (MLP) ndi 13.1% (MLP), motsatana, kusonyeza kuti kusiyana kwa magwiridwe antchito pakati pa mitundu kumaposa kukhudzidwa. Pakati pa mitundu ya DM, makina othandizira (SVM), mtengo wosankha (DT), ndi mitundu ya nkhalango yosasinthika (RF) adachita bwino kwambiri pakati pa amuna, pomwe mtundu wa LR udachita bwino kwambiri pakati pa akazi. AUROC ya mtundu wachikhalidwe ndi mitundu yonse ya SD inali yoposa 0.925 (k-pafupi ndi mnansi (KNN) mwa amuna), zomwe zikuwonetsa magwiridwe antchito abwino kwambiri pakusankha zitsanzo za ana azaka 1828. Pa seti yoyesera yakunja, panali kuchepa kwa magwiridwe antchito pakugawa malinga ndi kukhudzika, kulunjika, ndi AUROC poyerekeza ndi seti yoyesera yamkati. Kuphatikiza apo, kusiyana kwa kukhudzika ndi kulunjika pakati pa magwiridwe antchito akugawa a mitundu yabwino kwambiri ndi yoyipa kunayambira pa 10% mpaka 25% ndipo kunali kwakukulu kuposa kusiyana kwa seti yoyesera yamkati.
Kuzindikira komanso kudziwika bwino kwa mitundu yogawa deta poyerekeza ndi njira zachikhalidwe zokhala ndi nthawi yocheperako ya zaka 18. KNN k pafupi ndi mnansi, makina othandizira a SVM, LR logistic regression, mtengo wosankha wa DT, RF random forest, XGB XGBoost, MLP multilayer perceptron, njira yachikhalidwe ya CM.
Gawo loyamba mu kafukufukuyu linali kuyerekeza kulondola kwa ziwerengero za zaka za mano zomwe zapezeka kuchokera ku mitundu isanu ndi iwiri ya DM ndi zomwe zapezeka pogwiritsa ntchito njira yachikhalidwe yosinthira. MAE ndi RMSE zinayesedwa m'mayesero amkati mwa amuna ndi akazi, ndipo kusiyana pakati pa njira yachikhalidwe ndi chitsanzo cha DM kunayambira masiku 44 mpaka 77 a MAE ndi masiku 62 mpaka 88 a RMSE. Ngakhale njira yachikhalidwe inali yolondola pang'ono mu kafukufukuyu, n'kovuta kunena ngati kusiyana kochepa koteroko kuli ndi tanthauzo lachipatala kapena lothandiza. Zotsatirazi zikusonyeza kuti kulondola kwa kuwerengera zaka za mano pogwiritsa ntchito chitsanzo cha DM kuli kofanana ndi kwa njira yachikhalidwe. Kuyerekeza mwachindunji ndi zotsatira kuchokera ku maphunziro am'mbuyomu n'kovuta chifukwa palibe kafukufuku amene wayerekeza kulondola kwa zitsanzo za DM ndi njira zachikhalidwe zowerengera pogwiritsa ntchito njira yomweyo yolembera mano mu msinkhu womwewo monga momwe zinalili mu kafukufukuyu. Galibourg et al24 anayerekeza MAE ndi RMSE pakati pa njira ziwiri zachikhalidwe (Demirjian method25 ndi Willems method29) ndi zitsanzo 10 za DM mwa anthu aku France azaka 2 mpaka 24. Iwo adanenanso kuti mitundu yonse ya DM inali yolondola kwambiri kuposa njira zachikhalidwe, ndi kusiyana kwa zaka 0.20 ndi 0.38 mu MAE ndi zaka 0.25 ndi 0.47 mu RMSE poyerekeza ndi njira za Willems ndi Demirdjian, motsatana. Kusiyana pakati pa chitsanzo cha SD ndi njira zachikhalidwe zomwe zawonetsedwa mu kafukufuku wa Halibourg kumaganizira malipoti ambiri 30, 31, 32, 33 kuti njira ya Demirdjian simawerengera molondola zaka za mano m'magulu ena kupatula aku Canada aku France omwe kafukufukuyu adachokera. mu kafukufukuyu. Tai et al 34 adagwiritsa ntchito njira ya MLP kulosera zaka za mano kuchokera ku zithunzi za orthodontic zaku China za 1636 ndikuyerekeza kulondola kwake ndi zotsatira za njira ya Demirjian ndi Willems. Adanenanso kuti MLP ili ndi kulondola kwakukulu kuposa njira zachikhalidwe. Kusiyana pakati pa njira ya Demirdjian ndi njira yachikhalidwe ndi zaka <0.32, ndipo njira ya Willems ndi zaka 0.28, zomwe zikufanana ndi zotsatira za kafukufukuyu. Zotsatira za maphunziro am'mbuyomu24,34 zikugwirizananso ndi zotsatira za kafukufukuyu, ndipo kulondola kwa kuyerekezera zaka kwa chitsanzo cha DM ndi njira yachikhalidwe ndizofanana. Komabe, kutengera zotsatira zomwe zaperekedwa, tingathe kungoganiza mosamala kuti kugwiritsa ntchito mitundu ya DM poyesa zaka kungalowe m'malo mwa njira zomwe zilipo chifukwa chosowa maphunziro oyerekeza komanso ofotokozera am'mbuyomu. Maphunziro otsatira pogwiritsa ntchito zitsanzo zazikulu amafunika kuti atsimikizire zotsatira zomwe zapezeka mu kafukufukuyu.
Pakati pa maphunziro omwe adayesa kulondola kwa SD pakuyerekeza zaka za mano, ena adawonetsa kulondola kwakukulu kuposa kafukufuku wathu. Stepanovsky et al 35 adagwiritsa ntchito mitundu 22 ya SD pa ma radiograph a panoramic a anthu 976 aku Czech azaka zapakati pa 2.7 ndi 20.5 ndipo adayesa kulondola kwa mtundu uliwonse. Adayesa chitukuko cha mano okhazikika 16 apamwamba ndi otsika kumanzere pogwiritsa ntchito njira zogawa zomwe Moorrees et al 36 adapereka. MAE imayambira pa 0.64 mpaka 0.94 ndipo RMSE imayambira pa 0.85 mpaka 1.27, zomwe ndizolondola kwambiri kuposa mitundu iwiri ya DM yomwe idagwiritsidwa ntchito mu kafukufukuyu. Shen et al23 adagwiritsa ntchito njira ya Cameriere kuti ayese zaka za mano asanu ndi awiri okhazikika m'manja mwamanzere mwa anthu akum'mawa kwa China azaka zapakati pa 5 ndi 13 ndipo adayerekeza ndi zaka zomwe zidaganiziridwa pogwiritsa ntchito linear regression, SVM ndi RF. Adawonetsa kuti mitundu yonse itatu ya DM ili ndi kulondola kwakukulu poyerekeza ndi njira yachikhalidwe ya Cameriere. MAE ndi RMSE mu kafukufuku wa Shen zinali zochepa kuposa zomwe zili mu chitsanzo cha DM mu kafukufukuyu. Kulondola kwakukulu kwa maphunzirowa ndi Stepanovsky et al. 35 ndi Shen et al. 23 kungakhale chifukwa cha kuphatikizidwa kwa anthu achichepere mu zitsanzo zawo zophunzirira. Chifukwa kuyerekezera kwa zaka kwa ophunzira omwe ali ndi mano omwe akukula kumakhala kolondola kwambiri pamene chiwerengero cha mano chikuwonjezeka panthawi yokukula kwa mano, kulondola kwa njira yowunikira zaka zomwe zapezeka kungasokonezedwe pamene ophunzira omwe ali ndi kafukufukuyu ali aang'ono. Kuphatikiza apo, cholakwika cha MLP pakuyerekeza zaka ndi chochepa pang'ono kuposa cha SLP, zomwe zikutanthauza kuti MLP ndi yolondola kwambiri kuposa SLP. MLP imaonedwa kuti ndi yabwinoko pang'ono pakuyerekeza zaka, mwina chifukwa cha zigawo zobisika mu MLP38. Komabe, pali zosiyana ndi chitsanzo chakunja cha akazi (SLP 1.45, MLP 1.49). Kupeza kuti MLP ndi yolondola kwambiri kuposa SLP pakuyesa zaka kumafuna maphunziro ena obwerezabwereza.
Kugwira ntchito kwa chitsanzo cha DM ndi njira yachikhalidwe pamlingo wa zaka 18 kunayerekezeredwanso. Ma model onse a SD oyesedwa ndi njira zachikhalidwe pa seti yoyesera yamkati adawonetsa milingo yovomerezeka ya tsankho kwa chitsanzo cha zaka 18. Kukhudzidwa kwa amuna ndi akazi kunali kwakukulu kuposa 87.7% ndi 94.9%, motsatana, ndipo kudziwika kwake kunali kwakukulu kuposa 89.3% ndi 84.7%. AUROC ya ma model onse oyesedwa imaposa 0.925. Monga momwe tikudziwira, palibe kafukufuku amene adayesa kugwira ntchito kwa chitsanzo cha DM cha zaka 18 kutengera kukula kwa mano. Tikhoza kufananiza zotsatira za kafukufukuyu ndi kugwira ntchito kwa ma model a deep learning pa panoramic radiographs. Guo et al.15 adawerengera kugwira ntchito kwa CNN-based deep learning model ndi njira yamanja kutengera njira ya Demirjian ya msinkhu winawake. Kuzindikira ndi kulunjika kwa njira yogwiritsira ntchito pamanja kunali 87.7% ndi 95.5%, motsatana, ndipo kukhudzidwa ndi kulunjika kwa chitsanzo cha CNN kunapitirira 89.2% ndi 86.6%, motsatana. Anatsimikiza kuti mitundu yophunzirira mozama imatha kusintha kapena kuchita bwino kuposa kuwunika kwamanja pakugawa malire a zaka. Zotsatira za kafukufukuyu zinawonetsa magwiridwe antchito ofanana; Akukhulupirira kuti kugawa pogwiritsa ntchito mitundu ya DM kumatha kusintha njira zachikhalidwe zowerengera zaka. Pakati pa mitundu, DM LR inali chitsanzo chabwino kwambiri pankhani ya kukhudzidwa kwa chitsanzo cha amuna ndi kukhudzidwa ndi kulunjika kwa chitsanzo cha akazi. LR ili pa nambala yachiwiri pa kulunjika kwa amuna. Kuphatikiza apo, LR imaonedwa kuti ndi imodzi mwa mitundu ya DM35 yosavuta kugwiritsa ntchito ndipo ndi yosavuta komanso yovuta kuikonza. Kutengera ndi zotsatirazi, LR idaonedwa ngati chitsanzo chabwino kwambiri chogawa kwa ana azaka 18 mwa anthu aku Korea.
Ponseponse, kulondola kwa kuyerekezera zaka kapena kuchita bwino kwa magulu pa mayeso akunja kunali koipa kapena kotsika poyerekeza ndi zotsatira za mayeso amkati. Malipoti ena akusonyeza kuti kulondola kapena kugwira ntchito bwino kwa magulu kumachepa pamene kuyerekezera zaka kutengera anthu aku Korea kukugwiritsidwa ntchito kwa anthu aku Japan5,39, ndipo njira yofananayi idapezeka mu kafukufukuyu. Kuwonongeka kumeneku kudawonedwanso mu chitsanzo cha DM. Chifukwa chake, kuti muyesere molondola zaka, ngakhale mukugwiritsa ntchito DM pakusanthula, njira zochokera ku deta ya anthu akumaloko, monga njira zachikhalidwe, ziyenera kukondedwa5,39,40,41,42. Popeza sizikudziwika ngati mitundu yophunzirira mozama ingawonetse zomwe zikuchitika, maphunziro oyerekeza kulondola kwa magulu ndi magwiridwe antchito pogwiritsa ntchito njira zachikhalidwe, mitundu ya DM, ndi mitundu yophunzirira mozama pa zitsanzo zomwezo amafunika kuti atsimikizire ngati luntha lochita kupanga lingathe kuthana ndi kusiyana kwa mitundu kumeneku pakuyesa kwa zaka zochepa.
Tikuwonetsa kuti njira zachikhalidwe zitha kusinthidwa ndi kuyerekezera zaka kutengera chitsanzo cha DM mu machitidwe owerengera zaka za forensic ku Korea. Tinapezanso kuthekera kogwiritsa ntchito makina ophunzirira pakuwunika zaka za forensic. Komabe, pali zoletsa zomveka bwino, monga kuchuluka kosakwanira kwa ophunzira mu kafukufukuyu kuti adziwe bwino zotsatira zake, komanso kusowa kwa maphunziro am'mbuyomu kuti ayerekeze ndikutsimikizira zotsatira za kafukufukuyu. M'tsogolomu, maphunziro a DM ayenera kuchitidwa ndi zitsanzo zambiri komanso anthu osiyanasiyana kuti akonze momwe angagwiritsire ntchito poyerekeza ndi njira zachikhalidwe. Kuti atsimikizire kuthekera kogwiritsa ntchito luntha lochita kupanga kuti awerengere zaka mwa anthu ambiri, maphunziro amtsogolo amafunika kuti ayerekeze kulondola ndi magwiridwe antchito a DM ndi mitundu yophunzirira mozama ndi njira zachikhalidwe m'zitsanzo zomwezo.
Kafukufukuyu adagwiritsa ntchito zithunzi 2,657 zolembedwa kuchokera kwa akuluakulu aku Korea ndi Japan azaka zapakati pa 15 mpaka 23. Ma X-ray a ku Korea adagawidwa m'magulu 900 ophunzitsira (zaka 19.42 ± 2.65) ndi magulu 900 oyesera mkati (zaka 19.52 ± 2.59). Gulu lophunzitsira lidasonkhanitsidwa ku bungwe limodzi (Chipatala cha Seoul St. Mary), ndipo gulu loyesera lidasonkhanitsidwa m'mabungwe awiri (Chipatala cha Mano cha Seoul National University ndi Chipatala cha Mano cha Yonsei University). Tinasonkhanitsanso ma X-ray 857 kuchokera ku deta ina yochokera kwa anthu (Iwate Medical University, Japan) kuti ayesedwe kunja. Ma X-ray a anthu aku Japan (zaka 19.31 ± 2.60) adasankhidwa ngati gulu loyesera kunja. Deta idasonkhanitsidwa mobwerezabwereza kuti iwunike magawo akukula kwa mano pa ma X-ray a panoramic omwe adatengedwa panthawi ya chithandizo cha mano. Deta yonse yosonkhanitsidwa sinali yodziwika kupatula jenda, tsiku lobadwa ndi tsiku la X-ray. Njira zophatikizira ndi zochotsera zinali zofanana ndi maphunziro omwe adasindikizidwa kale 4, 5. Zaka zenizeni za chitsanzocho zinawerengedwa pochotsa tsiku lobadwa kuchokera tsiku lomwe chithunzi cha x-ray chinatengedwa. Gulu la zitsanzolo linagawidwa m'magulu azaka zisanu ndi zinayi. Kugawa kwa zaka ndi kugonana kwawonetsedwa mu Table 3. Kafukufukuyu adachitika motsatira Chilengezo cha Helsinki ndipo adavomerezedwa ndi Institutional Review Board (IRB) ya Seoul St. Mary's Hospital of the Catholic University of Korea (KC22WISI0328). Chifukwa cha kapangidwe ka kafukufukuyu, chilolezo chodziwitsidwa sichinapezeke kwa odwala onse omwe akuyesedwa ndi x-ray kuti achiritsidwe. Seoul Korea University St. Mary's Hospital (IRB) idasiya kufunikira kwa chilolezo chodziwitsidwa.
Magawo a kukula kwa mano achiwiri ndi achitatu a bimaxillary adayesedwa malinga ndi zofunikira za Demircan25. Dzino limodzi lokha linasankhidwa ngati mtundu womwewo wa dzino unapezeka mbali yakumanzere ndi yakumanja ya nsagwada iliyonse. Ngati mano ofanana mbali zonse ziwiri anali pa magawo osiyanasiyana akukula, dzino lomwe lili ndi gawo lotsika lakukula linasankhidwa kuti liwerengere kusatsimikizika kwa zaka zomwe zaganiziridwa. Ma X-ray osankhidwa mwachisawawa zana limodzi kuchokera ku seti yophunzitsira adawerengedwa ndi owonera awiri odziwa bwino ntchito kuti ayesere kudalirika kwa owonera pambuyo poyesa kukhazikika kwa mano kuti adziwe nthawi yokhwima kwa mano. Kudalirika kwa owonera mkati kunayesedwa kawiri pakatha miyezi itatu ndi wowonera woyamba.
Gawo la kugonana ndi kukula kwa malo achiwiri ndi achitatu a nsagwada iliyonse mu seti yophunzitsira linayesedwa ndi woyang'anira wamkulu wophunzitsidwa ndi mitundu yosiyanasiyana ya DM, ndipo zaka zenizeni zinayikidwa ngati mtengo wofunikira. Mitundu ya SLP ndi MLP, yomwe imagwiritsidwa ntchito kwambiri mu kuphunzira kwa makina, inayesedwa motsutsana ndi ma algorithms a regression. Chitsanzo cha DM chimaphatikiza ntchito zolunjika pogwiritsa ntchito magawo okukula a mano anayi ndikusakaniza deta iyi kuti iyerekeze zaka. SLP ndi netiweki yosavuta kwambiri ya mitsempha ndipo ilibe zigawo zobisika. SLP imagwira ntchito kutengera kutumiza pakati pa ma node. Chitsanzo cha SLP mu regression chimafanana ndi masamu ndi multiple linear regression. Mosiyana ndi chitsanzo cha SLP, chitsanzo cha MLP chili ndi zigawo zambiri zobisika zokhala ndi ntchito zoyambitsa nonlinear. Kuyesera kwathu kunagwiritsa ntchito gawo lobisika lokhala ndi ma node 20 obisika okha okhala ndi ntchito zoyambitsa nonlinear. Gwiritsani ntchito gradient descent ngati njira yokonzera ndi MAE ndi RMSE ngati ntchito yotayika kuti muphunzitse chitsanzo chathu cha makina ophunzirira. Chitsanzo chabwino kwambiri chobwezeretsa chinagwiritsidwa ntchito ku mayeso amkati ndi akunja ndipo zaka za mano zinayerekezeredwa.
Njira yogawa deta idapangidwa yomwe imagwiritsa ntchito kukhwima kwa mano anayi pa seti yophunzitsira kuti idziwike ngati chitsanzo chili ndi zaka 18 kapena ayi. Kuti tipange chitsanzochi, tinapeza ma algorithm asanu ndi awiri oyimira makina ophunzirira6,43: (1) LR, (2) KNN, (3) SVM, (4) DT, (5) RF, (6) XGBoost, ndi (7) MLP. LR ndi imodzi mwa ma algorithm ogawa deta omwe amagwiritsidwa ntchito kwambiri44. Ndi njira yowunikira yowunikira yomwe imagwiritsa ntchito regression kuti ilosere kuthekera kwa deta kukhala ya gulu linalake kuyambira 0 mpaka 1 ndikugawa detayo ngati ya gulu lomwe lingakhalepo kutengera kuthekera kumeneku; makamaka imagwiritsidwa ntchito pogawa deta ya binary. KNN ndi imodzi mwa ma algorithm osavuta kwambiri ophunzirira makina45. Ikapatsidwa deta yatsopano yolowera, imapeza deta ya k pafupi ndi seti yomwe ilipo kenako nkuwagawa m'kalasi yokhala ndi ma frequency apamwamba kwambiri. Timayika 3 pa chiwerengero cha anansi omwe akuganiziridwa (k). SVM ndi njira yowunikira yomwe imakulitsa mtunda pakati pa magulu awiri pogwiritsa ntchito ntchito ya kernel kuti ikule malo olunjika kukhala malo osakhala olunjika otchedwa fields46. Pa chitsanzo ichi, timagwiritsa ntchito bias = 1, power = 1, ndi gamma = 1 ngati hyperparameters ya polynomial kernel. DT yagwiritsidwa ntchito m'magawo osiyanasiyana ngati algorithm yogawa deta yonse m'magulu angapo poyimira malamulo osankha mu kapangidwe ka mtengo47. Chitsanzocho chimakonzedwa ndi chiwerengero chochepa cha zolemba pa node ya 2 ndipo chimagwiritsa ntchito Gini index ngati muyeso wa khalidwe. RF ndi njira yophatikiza yomwe imaphatikiza ma DT angapo kuti ikonze magwiridwe antchito pogwiritsa ntchito njira yolumikizirana ya bootstrap yomwe imapanga gulu lofooka la chitsanzo chilichonse mwa kujambula mwachisawawa zitsanzo za kukula komweko kangapo kuchokera ku dataset yoyambirira48. Tinagwiritsa ntchito mitengo 100, kuya kwa mitengo 10, kukula kwa node imodzi kocheperako, ndi Gini admixture index ngati njira yolekanitsira node. Kugawa deta yatsopano kumatsimikiziridwa ndi mavoti ambiri. XGBoost ndi algorithm yomwe imaphatikiza njira zolimbikitsira pogwiritsa ntchito njira yomwe imatenga ngati deta yophunzitsira cholakwika pakati pa zenizeni ndi zomwe zanenedweratu za chitsanzo cham'mbuyomu ndikuwonjezera cholakwikacho pogwiritsa ntchito gradients49. Ndi njira yogwiritsidwa ntchito kwambiri chifukwa cha magwiridwe ake abwino komanso kugwiritsa ntchito bwino zinthu, komanso kudalirika kwambiri ngati ntchito yokonza zinthu mopitirira muyeso. Chitsanzochi chili ndi mawilo othandizira 400. MLP ndi netiweki ya mitsempha momwe ma perceptron amodzi kapena angapo amapanga zigawo zingapo ndi gawo limodzi kapena angapo obisika pakati pa zigawo zolowetsa ndi zotulutsa38. Pogwiritsa ntchito izi, mutha kuchita magulu osakhala a mzere pomwe mukawonjezera gawo lolowera ndikupeza phindu la zotsatira, phindu la zotsatira lomwe lanenedweratu limayerekezeredwa ndi phindu lenileni la zotsatira ndipo cholakwikacho chimafalikira mmbuyo. Tinapanga gawo lobisika ndi ma neuron 20 obisika mu gawo lililonse. Chitsanzo chilichonse chomwe tidapanga chidagwiritsidwa ntchito ku ma seti amkati ndi akunja kuti tiyese magwiridwe antchito a gulu powerengera kukhudzika, kutsimikizika, PPV, NPV, ndi AUROC. Kukhudzidwa kumatanthauzidwa ngati chiŵerengero cha chitsanzo chomwe chikuyerekezeredwa kukhala ndi zaka 18 kapena kupitirira apo ku chitsanzo chomwe chikuyerekezeredwa kukhala ndi zaka 18 kapena kupitirira apo. Kutsimikizika ndi gawo la zitsanzo zomwe zili pansi pa zaka 18 ndi zomwe zikuyerekezeredwa kukhala pansi pa zaka 18.
Magawo a mano omwe adawunikidwa mu seti yophunzitsira adasinthidwa kukhala magawo a manambala kuti ayesedwe ziwerengero. Kubwerezabwereza kwa mzere ndi logistic kwa multivariate kunachitika kuti apange zitsanzo zolosera za kugonana kulikonse ndikupeza njira zobwerezabwereza zomwe zingagwiritsidwe ntchito kuyerekeza zaka. Tinagwiritsa ntchito njirazi kuyerekeza zaka za mano pamaseti oyesera amkati ndi akunja. Gome 4 likuwonetsa mitundu yobwerezabwereza ndi magulu omwe agwiritsidwa ntchito mu kafukufukuyu.
Kudalirika kwa mkati ndi mkati mwa owonera kunawerengedwa pogwiritsa ntchito ziwerengero za Cohen's kappa. Kuti tiyese kulondola kwa DM ndi zitsanzo zachikhalidwe za regression, tinawerengera MAE ndi RMSE pogwiritsa ntchito zaka zoyerekezeredwa komanso zenizeni za mayeso amkati ndi akunja. Zolakwika izi zimagwiritsidwa ntchito nthawi zambiri poyesa kulondola kwa maulosi a chitsanzo. Cholakwika chikakhala chaching'ono, kulondola kwa kulosera kumakwera24. Yerekezerani MAE ndi RMSE ya mayeso amkati ndi akunja omwe adawerengedwa pogwiritsa ntchito DM ndi regression yachikhalidwe. Kuchita bwino kwa magulu a zaka 18 mu ziwerengero zachikhalidwe kunayesedwa pogwiritsa ntchito tebulo la 2 × 2. Kuzindikira kowerengedwa, kutsimikizika, PPV, NPV, ndi AUROC ya seti yoyesera kunayerekezeredwa ndi mitengo yoyesedwa ya chitsanzo cha gulu la DM. Deta imafotokozedwa ngati mean ± standard deviation kapena number (%) kutengera mawonekedwe a deta. Ma P-side-side values <0.05 amaonedwa kuti ndi ofunika kwambiri. Kusanthula konse kwa ziwerengero kumachitika pogwiritsa ntchito SAS version 9.4 (SAS Institute, Cary, NC). Chitsanzo cha DM regression chinagwiritsidwa ntchito mu Python pogwiritsa ntchito Keras50 2.2.4 backend ndi Tensorflow51 1.8.0 makamaka pa ntchito zamasamu. Chitsanzo cha DM classification chinagwiritsidwa ntchito mu Waikato Knowledge Analysis Environment ndi nsanja yowunikira ya Konstanz Information Miner (KNIME) 4.6.152.
Olembawo akuvomereza kuti deta yothandizira zomwe zapezeka mu kafukufukuyu ingapezeke m'nkhani ndi m'zinthu zina zowonjezera. Ma data omwe apangidwa ndi/kapena kufufuzidwa panthawi ya kafukufukuyu akupezeka kwa wolemba woyenerera ngati pakufunika kutero.
Ritz-Timme, S. ndi ena. Kuwunika zaka: zamakono kuti zikwaniritse zofunikira zenizeni za ntchito ya zamalamulo. dziko lonse lapansi. J. Mankhwala azamalamulo. 113, 129–136 (2000).
Schmeling, A., Reisinger, W., Geserik, G., ndi Olze, A. Mkhalidwe wamakono wa kuwunika zaka za anthu omwe ali ndi moyo kuti apeze milandu. Forensic. medicine. Pathology. 1, 239–246 (2005).
Pan, J. et al. Njira yosinthidwa yowunikira zaka za mano za ana azaka zapakati pa 5 ndi 16 kum'mawa kwa China. zachipatala. Kafukufuku wa pakamwa. 25, 3463–3474 (2021).
Lee, SS etc. Kuchuluka kwa nthawi ya chitukuko cha molars yachiwiri ndi yachitatu mwa anthu aku Korea ndi momwe imagwiritsidwira ntchito poyesa zaka za akatswiri azachipatala. Internationality. J. Legal medicine. 124, 659–665 (2010).
Oh, S., Kumagai, A., Kim, SY ndi Lee, SS Kulondola kwa kuyerekezera zaka ndi kuyerekezera malire a zaka 18 kutengera kukula kwa molars yachiwiri ndi yachitatu mwa anthu aku Korea ndi aku Japan. PLoS ONE 17, e0271247 (2022).
Kim, JY, ndi ena. Kusanthula deta yochokera ku kuphunzira kwa makina musanachite opaleshoni kungathe kulosera zotsatira za chithandizo cha opaleshoni ya tulo mwa odwala omwe ali ndi OSA. sayansi. Lipoti 11, 14911 (2021).
Han, M. ndi ena. Kuyerekeza zaka molondola kuchokera ku kuphunzira kwa makina ndi kapena popanda kulowererapo kwa anthu? mayiko. J. Mankhwala azamalamulo. 136, 821–831 (2022).
Khan, S. ndi Shaheen, M. Kuchokera ku Kufufuza Deta mpaka Kufufuza Deta. J.Information. sayansi. https://doi.org/10.1177/01655515211030872 (2021).
Khan, S. ndi Shaheen, M. WisRule: Njira Yoyamba Yoganizira za Kugwira Ntchito ndi Malamulo a Kugwirizanitsa. J.Information. sayansi. https://doi.org/10.1177/01655515221108695 (2022).
Shaheen M. ndi Abdullah U. Karm: Kufufuza deta mwachikhalidwe kutengera malamulo ogwirizana ndi nkhani. werengerani. Matt. pitirizani. 68, 3305–3322 (2021).
Muhammad M., Rehman Z., Shaheen M., Khan M. ndi Habib M. Kuzindikira kufanana kwa mawu pogwiritsa ntchito deta yolembedwa. inform. technologies. control. https://doi.org/10.5755/j01.itc.49.4.27118 (2020).
Tabish, M., Tanoli, Z., ndi Shahin, M. Dongosolo lozindikira zochitika mu makanema amasewera. multimedia. Zida Mapulogalamu https://doi.org/10.1007/s11042-021-10519-6 (2021).
Halabi, SS ndi ena. RSNA Machine Learning Challenge mu Ana Bone Age. Radiology 290, 498–503 (2019).
Li, Y. ndi ena. Kuyerekeza zaka kuchokera ku X-ray ya m'chiuno pogwiritsa ntchito kuphunzira mozama. EURO. radiation. 29, 2322–2329 (2019).
Guo, YC, ndi ena. Kugawa zaka molondola pogwiritsa ntchito njira zamanja ndi maukonde ozama a mitsempha yolumikizana kuchokera ku zithunzi zowonetsera zolemba. dziko lonse lapansi. J. Mankhwala azamalamulo. 135, 1589–1597 (2021).
Alabama Dalora ndi ena. Kuyerekeza zaka za mafupa pogwiritsa ntchito njira zosiyanasiyana zophunzirira makina: kuwunikanso mabuku mwadongosolo komanso kusanthula meta. PLoS ONE 14, e0220242 (2019).
Du, H., Li, G., Cheng, K., ndi Yang, J. Kuyerekeza zaka za anthu aku Africa America ndi aku China kutengera kuchuluka kwa malo oyambira a pulp chamber pogwiritsa ntchito cone-beam computed tomography. internationality. J. Legal medicine. 136, 811–819 (2022).
Kim S., Lee YH, Noh YK, Park FK ndi Oh KS Kupeza magulu azaka za anthu amoyo pogwiritsa ntchito zithunzi zochokera ku luntha lochita kupanga za molars yoyamba. sayansi. Lipoti 11, 1073 (2021).
Stern, D., Payer, C., Giuliani, N., ndi Urschler, M. Kuwerengera zaka zokha ndi kugawa zaka zambiri kuchokera ku data ya MRI ya multivariate. IEEE J. Biomed. Zidziwitso Zaumoyo. 23, 1392–1403 (2019).
Cheng, Q., Ge, Z., Du, H. ndi Li, G. Kuyerekeza zaka kutengera kugawa kwa chipinda cha pulp cha 3D cha molars yoyamba kuchokera ku cone beam computed tomography pophatikiza kuphunzira kwakuya ndi magulu a level. internationality. J. Legal medicine. 135, 365–373 (2021).
Wu, WT, ndi ena. Kufufuza deta mu deta yayikulu yachipatala: ma database wamba, masitepe, ndi njira zitsanzo. World. medicine. resource. 8, 44 (2021).
Yang, J. ndi ena. Chiyambi cha Ma Database Azachipatala ndi Ukadaulo Wogwiritsa Ntchito Deta mu Nthawi Yaikulu ya Deta. J. Avid. Mankhwala Oyambira. 13, 57–69 (2020).
Shen, S. ndi ena. Njira ya Camerer yowerengera zaka za dzino pogwiritsa ntchito makina ophunzirira. BMC Oral Health 21, 641 (2021).
Galliburg A. ndi ena. Kuyerekeza njira zosiyanasiyana zophunzirira makina poneneratu zaka za mano pogwiritsa ntchito njira ya Demirdjian staging. internationality. J. Legal medicine. 135, 665–675 (2021).
Demirdjian, A., Goldstein, H. ndi Tanner, JM Njira yatsopano yowunikira zaka za mano. snort. biology. 45, 211–227 (1973).
Landis, JR, ndi Koch, GG Muyeso wa mgwirizano wa owonera pa deta yamagulu. Biometrics 33, 159–174 (1977).
Bhattacharjee S, Prakash D, Kim C, Kim HK ndi Choi HK. Kusanthula kwa kapangidwe kake, kapangidwe kake ndi ziwerengero za kujambula kwa maginito kogwiritsa ntchito njira zanzeru zopangira kusiyanitsa zotupa zoyambirira za muubongo. Chidziwitso chaumoyo. gwero. https://doi.org/10.4258/hir.2022.28.1.46 (2022).
Nthawi yotumizira: Januwale-04-2024