Tinokutendai nekushanyira Nature.com. Shanduro yebrowser yauri kushandisa ine rutsigiro rushoma rweCSS. Kuti uwane mhedzisiro yakanaka, tinokurudzira kushandisa shanduro itsva yebrowser yako (kana kudzima compatibility mode muInternet Explorer). Panguva ino, kuti tive nechokwadi chekuti rutsigiro rwuripo, tiri kuratidza webhusaiti iyi pasina styling kana JavaScript.
Mazino anoonekwa sechiratidzo chakarurama chezera remuviri wemunhu uye anowanzo shandiswa mukuongorora makore ekuongorora. Chinangwa chedu ndechekusimbisa fungidziro yezera remazino rakavakirwa pakuchera data nekuenzanisa kururama kwekufungidzira uye mashandiro ekupatsanura kwechikamu chemakore gumi nemasere nenzira dzechinyakare uye fungidziro yezera rekuchera data. Huwandu hwema panoramic radiographs 2657 akaunganidzwa kubva kuvagari vekuKorea neJapan vane makore gumi nemashanu kusvika makumi maviri nematatu. Akakamurwa kuita seti yekudzidzira, imwe neimwe iine ma radiographs mazana mapfumbamwe ekuKorea, uye seti yebvunzo yemukati ine ma radiographs eJapan mazana masere nemakumi mashanu nemanomwe. Takaenzanisa kururama kwekuisa muzvikamu uye kushanda kwenzira dzechinyakare neseti dzebvunzo dzema modhi ekuchera data. Kururama kwenzira yechinyakare pa seti yebvunzo yemukati kwakakwira zvishoma pane kwemuenzaniso wekuchera data, uye musiyano mudiki (avhareji kukanganisa kwakakwana <0.21 years, root mean square error <0.24 years). Kushanda muzvikamu kwechikamu chemakore gumi nemasere kwakafananawo pakati penzira dzechinyakare nemodhi dzekuchera data. Saka, nzira dzechinyakare dzinogona kutsiviwa nemaitiro ekutsvaga data pakuita ongororo yezera rekuongorora uchishandisa kukura kwemolars yechipiri neyechitatu muvechidiki vekuKorea nevechidiki.
Kuongororwa kwezera remazino kunoshandiswa zvakanyanya mukurapa kwechiremba wezvekurapa uye mukurapa mazino evana. Kunyanya, nekuda kwehukama hwakakura pakati pezera renguva nekukura kwemazino, kuongororwa kwezera nematanho ekukura kwemazino chinhu chakakosha pakuongorora zera revana nevechiri kuyaruka1,2,3. Zvisinei, kune vechidiki, kuongorora zera remazino zvichibva pakukura kwemazino kune miganhu yaro nekuti kukura kwemazino kwava kuda kupera, kunze kwemazino echitatu. Chinangwa chepamutemo chekuona zera revechidiki nevechiri kuyaruka ndechekupa fungidziro dzakarurama uye humbowo hwesainzi hwekuti vasvika zera rekukura here. Mukurapa-mutemo kwevechiri kuyaruka nevechidiki muKorea, zera rakafungidzirwa uchishandisa nzira yaLee, uye chiyero chepamutemo chemakore gumi nemasere chakafanotaurwa zvichibva padata rakataurwa naOh et al 5.
Kudzidza kwemuchina rudzi rwehungwaru hwekugadzira (AI) hunodzokorora kudzidza nekuronga huwandu hwakawanda hwedata, kugadzirisa matambudziko ega, uye kutungamira kuronga data. Kudzidza kwemuchina kunogona kuwana mapatani akavanzika anobatsira muhuwandu hwakawanda hwedata6. Kusiyana neizvi, nzira dzekare, dzinoda basa rakawanda uye dzinotora nguva, dzinogona kunge dziine zvipingamupinyi pakubata nehuwandu hwakawanda hwedata rakaoma kunzwisisa rinonetsa kugadzirisa nemaoko7. Nokudaro, zvidzidzo zvakawanda zvakaitwa munguva pfupi yapfuura uchishandisa matekinoroji emakombiyuta emazuva ano kuderedza zvikanganiso zvevanhu uye kugadzirisa data remadimensional zvinobudirira8,9,10,11,12. Kunyanya, kudzidza kwakadzama kwave kuchishandiswa zvakanyanya mukuongorora mifananidzo yekurapa, uye nzira dzakasiyana-siyana dzekuongorora zera nekuongorora otomatiki maradiographs dzakashumwa kuti dzinovandudza kururama uye kushanda kwekufungidzira zera13,14,15,16,17,18,19,20. Semuenzaniso, Halabi et al 13 vakagadzira algorithm yekudzidza kwemuchina yakavakirwa pa convolutional neural networks (CNN) yekufungidzira zera remapfupa vachishandisa radiographs dzemaoko evana. Chidzidzo ichi chinoratidza modhi inoshandisa kudzidza kwemuchina kumifananidzo yekurapa uye inoratidza kuti nzira idzi dzinogona kuvandudza kunyatsoongorora. Li et al14 vakafungidzira zera kubva pamifananidzo yeX-ray yepachiuno vachishandisa CNN yekudzidza kwakadzika uye vakaenzanisa nemigumisiro yekudzoka vachishandisa kufungidzira kwechikamu che ossification. Vakaona kuti modhi yeCNN yekudzidza kwakadzika yakaratidza mashandiro akafanana ekufungidzira zera semuenzaniso wekare wekudzoka. Chidzidzo chaGuo et al. [15] chakaongorora mashandiro eCNN ekugadzirisa zera zvichibva pamifananidzo yemazino, uye mhedzisiro yemodhi yeCNN yakaratidza kuti vanhu vakapfuura mashandiro ayo ekupatsanura zera.
Zvidzidzo zvakawanda pamusoro pekufungidzira zera uchishandisa machine learning zvinoshandisa nzira dzekudzidza zvakadzama13,14,15,16,17,18,19,20. Kufungidzira zera zvichibva pakudzidza zvakadzama kunonzi kwakarurama kupfuura nzira dzechinyakare. Zvisinei, nzira iyi haipe mukana wekuratidza hwaro hwesainzi hwekufungidzira zera, senge zviratidzo zvezera zvinoshandiswa mukufungidzira. Kune gakava repamutemo pamusoro pekuti ndiani anoita ongororo. Nokudaro, kufungidzira zera zvichibva pakudzidza zvakadzama kwakaoma kugamuchirwa nevatungamiriri nevatongi. Kuchera data (DM) inzira inogona kuwana kwete chete ruzivo rwunotarisirwa asiwo rusingatarisirwi senzira yekuwana hukama hunobatsira pakati pehuwandu hwakawanda hwedata6,21,22. Kudzidza kwemuchina kunowanzo shandiswa mukuchera data, uye zvese kuchera data nekudzidza kwemuchina zvinoshandisa ma algorithms akafanana kuti zviwane mapatani mudata. Kufungidzira zera uchishandisa kukura kwemazino kunobva pakuongorora kwemuongorori kwekukura kwemazino anotarisirwa, uye kuongorora uku kunoratidzwa sedanho rezino rega rega rakananga. DM inogona kushandiswa kuongorora hukama pakati pedanho rekuongorora mazino nezera chairo uye ine mukana wekutsiva ongororo yetsika yenhamba. Saka, kana tikashandisa matekiniki eDM pakuongorora zera, tinogona kushandisa machine learning mukuongorora zera re forensic tisinganetseki nezvemhosva yepamutemo. Zvidzidzo zvakati wandei zvakaburitswa pamusoro pedzimwe nzira dzechinyakare dzinoshandiswa nemaoko mukuita kwe forensic uye nzira dzakavakirwa paEBM pakuona zera remazino. Shen et al23 vakaratidza kuti modhi yeDM yakarurama kupfuura fomura yechinyakare yeCamerer. Galibourg et al24 vakashandisa nzira dzakasiyana dzeDM kufanotaura zera zvichienderana neDemirdjian criterion25 uye mhedzisiro yakaratidza kuti nzira yeDM yakabudirira kupfuura nzira dzaDemirdjian naWillems mukufungidzira zera revanhu vekuFrance.
Kuti tifungidzire zera remazino revechidiki vekuKorea nevechidiki, nzira yaLee 4 inoshandiswa zvakanyanya mukuita kwekuongorora mazino muKorea. Nzira iyi inoshandisa ongororo yechinyakare yenhamba (yakadai sekudzokorora kwakawanda) kuongorora hukama huripo pakati pevanhu vekuKorea nezera renguva. Muchidzidzo ichi, nzira dzekufungidzira zera dzakawanikwa uchishandisa nzira dzechinyakare dzenhamba dzinotsanangurwa se "nzira dzechinyakare." Nzira yaLee inzira yechinyakare, uye kururama kwayo kwakasimbiswa naOh et al. 5; zvisinei, kushanda kwekufungidzira zera zvichibva pamuenzaniso weDM mukuita kwekuongorora mazino muKorea kuchiri kusava nechokwadi. Chinangwa chedu chaive chekusimbisa nesainzi kushanda kwekufungidzira zera zvichibva pamuenzaniso weDM. Chinangwa chechidzidzo ichi chaive (1) kuenzanisa kururama kwemamodeli maviri eDM mukufungidzira zera remazino uye (2) kuenzanisa mashandiro ekupatsanura emamodeli manomwe eDM pazera remakore gumi nemasere neaya akawanikwa achishandisa nzira dzechinyakare dzenhamba Kukura kwemolars yechipiri neyechitatu mushaya dzese.
Nzira uye kutsauka kwakajairwa kwezera renguva zvichienderana nedanho uye rudzi rwezino zvinoratidzwa online muSupplementary Table S1 (seti yekudzidzira), Supplementary Table S2 (seti yekuedza kwemukati), uye Supplementary Table S3 (seti yekuedza kwekunze). Makoshero ekappa ekuvimbika kwemukati nemukati memuoni akawanikwa kubva museti yekudzidzira aive 0.951 uye 0.947, zvichiteerana. Makoshero eP uye 95% confidence intervals yekappa values anoratidzwa mu online supplementary table S4. Makoshero ekappa akadudzirwa se "anenge akakwana", zvichienderana nezvinodiwa naLandis naKoch26.
Pakuenzanisa mean absolute error (MAE), nzira yechinyakare inopfuura zvishoma DM model yevanhurume nevanhukadzi vese uye mu external male test set, kunze kwe multilayer perceptron (MLP). Musiyano uripo pakati pe traditional model neDM model pa internal MAE test set waive makore 0.12–0.19 kuvarume uye makore 0.17–0.21 kuvakadzi. Kune external test battery, mutsauko wacho mudiki (0.001–0.05 kuvarume uye makore 0.05–0.09 kuvakadzi). Pamusoro pezvo, root mean square error (RMSE) yakaderera zvishoma pane nzira yechinyakare, nemusiyano mudiki (0.17–0.24, 0.2–0.24 ye internal test set yevarume, uye 0.03–0.07, 0.04–0.08 ye external test set). ). MLP inoratidza kushanda kuri nani zvishoma pane Single Layer Perceptron (SLP), kunze kwekunge iri external test set yevakadzi. Kune MAE neRMSE, bvunzo dzekunze dzakakwira kupfuura bvunzo dzemukati dzevanhurume nevanhukadzi vese nemamodheru. MAE neRMSE dzese dzinoratidzwa muTafura 1 neMufananidzo 1.
MAE neRMSE yemamodheru ekare uye ekugadzirisa data. Chikanganiso chepakati cheMAE, chikanganiso chepakati chepakati cheRMSE, perceptron SLP ine layer imwe chete, perceptron MLP ine layer yakawanda, nzira yeCM yechinyakare.
Kushanda kwezvikamu (nekuchekwa kwemakore gumi nemasere) kwemamodheru echinyakare neDM kwakaratidzwa maererano nekunzwa, kujeka, kukosha kwekufungidzira kwakanaka (PPV), kukosha kwekufungidzira kusina kunaka (NPV), uye nzvimbo iri pasi pemuchina wekugamuchira (AUROC) 27 (Tafura 2, Mufananidzo 2 uye Mufananidzo Wekuwedzera 1 online). Nezvekunzwa kwebhatiri rekuyedza kwemukati, nzira dzechinyakare dzakashanda zvakanyanya pakati pevarume uye dzakaipa pakati pevakadzi. Zvisinei, musiyano mukuita kwezvikamu pakati penzira dzechinyakare neSD i9.7% yevarume (MLP) uye 2.4% chete yevakadzi (XGBoost). Pakati pemamodheru eDM, logistic regression (LR) yakaratidza kunzwisisika kuri nani muvanhurume nevarume. Nezvekunyatsojeka kweseti yekuyedza kwemukati, zvakaonekwa kuti mamodheru mana eSD akaita zvakanaka muvanhurume, nepo modheru yechinyakare yakaita zviri nani muvanhukadzi. Musiyano mukuita kwezvikamu kwevarume nevakadzi i13.3% (MLP) uye 13.1% (MLP), zvichiteerana, zvichiratidza kuti musiyano mukuita kwezvikamu pakati pemamodheru unopfuura kunzwisisika. Pakati pemamodeli eDM, mamodheru ekutsigira vector (SVM), decision tree (DT), uye mamodheru emusango asingatarisirwi (RF) akaita zvakanaka pakati pevarume, nepo mamodheru eLR akaita zvakanaka pakati pevakadzi. AUROC yemuenzaniso wekare uye mamodheru ese eSD aive akakura kupfuura 0.925 (k-pedyo nepedyo (KNN) muvarume), zvichiratidza kushanda zvakanaka mukusarudza masampuli evana vane makore 1828. Kune seti yebvunzo yekunze, paive nekudzikira kwekuita kwezvikamu maererano nekunzwa, kujeka uye AUROC zvichienzaniswa neseti yebvunzo yemukati. Uyezve, musiyano mukunzwa uye kujeka pakati pekuita kwezvikamu zvemamodheru akanakisa neakaipa waive kubva pa10% kusvika 25% uye waive wakakura kupfuura musiyano uri museti yebvunzo yemukati.
Kunzwisisa uye kujeka kwemamodheru ekuisa data muzvikamu zvichienzaniswa nenzira dzechinyakare dzine cutoff yemakore gumi nemasere. KNN k muvakidzani wepedyo, SVM support vector machine, LR logistic regression, DT decision tree, RF random forest, XGB XGBoost, MLP multilayer perceptron, traditional CM method.
Danho rekutanga muchidzidzo ichi raive rekuenzanisa kururama kwefungidziro yezera remazino yakawanikwa kubva kumamodeli manomwe eDM neaya akawanikwa achishandisa regression yechinyakare. MAE neRMSE zvakaongororwa mumaseti ebvunzo dzemukati mevanhurume nevanhukadzi, uye musiyano uripo pakati penzira yechinyakare nemuenzaniso weDM waive kubva pamazuva makumi mana nemana kusvika makumi manomwe nemanomwe eMAE uye kubva pamazuva makumi matanhatu nemaviri kusvika makumi masere nemasere eRMSE. Kunyangwe nzira yechinyakare yaive yakanyatsojeka muchidzidzo ichi, zvakaoma kugumisa kana musiyano mudiki wakadaro uine kukosha kwekiriniki kana kwekushanda. Mhedzisiro iyi inoratidza kuti kururama kwefungidziro yezera remazino uchishandisa muenzaniso weDM kwakafanana nekwenzira yechinyakare. Kuenzanisa zvakananga nemigumisiro kubva muzvidzidzo zvekare kwakaoma nekuti hapana chidzidzo chakaenzanisa kururama kwemamodeli eDM nenzira dzechinyakare dzekuverenga uchishandisa nzira imwechete yekunyora mazino muzera rimwe chete semuchidzidzo ichi. Galibourg et al24 vakaenzanisa MAE neRMSE pakati penzira mbiri dzechinyakare (Demirjian method25 naWillems method29) uye mamodeli gumi eDM muvanhu vekuFrance vane makore maviri kusvika makumi maviri nemana. Vakashuma kuti mamodheru ese eDM aive akarurama kupfuura nzira dzechinyakare, nemusiyano wemakore 0.20 ne0.38 muMAE uye makore 0.25 ne0.47 muRMSE zvichienzaniswa nenzira dzeWillems neDemirdjian, zvichiteerana. Musiyano uripo pakati peSD model nenzira dzechinyakare dzakaratidzwa muchidzidzo cheHalibourg unotora mishumo yakawanda30,31,32,33 yekuti nzira yeDemirdjian haifungidzire zera remazino nemazvo muvanhu vasiri vekuFrance vekuCanada kwakavakirwa chidzidzo ichi. muchidzidzo ichi. Tai et al 34 vakashandisa MLP algorithm kufanotaura zera remazino kubva pamifananidzo ye1636 yeChinese orthodontic uye vakaenzanisa kururama kwayo nemhedzisiro yenzira yeDemirjian neWillems. Vakashuma kuti MLP ine kururama kwakanyanya kupfuura nzira dzechinyakare. Musiyano uripo pakati penzira yeDemirdjian nenzira yechinyakare i<0.32 makore, uye nzira yeWillems imakore 0.28, izvo zvakafanana nemhedzisiro yechidzidzo ichi. Mhedzisiro yezvidzidzo izvi zvekare24,34 inoenderanawo nemhedzisiro yechidzidzo ichi, uye kururama kwekufungidzira zera kwemuenzaniso weDM nenzira yechinyakare zvakafanana. Zvisinei, zvichibva pane zvabuda, tinogona kungogumisa nekungwarira kuti kushandiswa kwemuenzaniso weDM kufungidzira zera kunogona kutsiva nzira dziripo nekuda kwekushaikwa kwezvidzidzo zvekare zvekuenzanisa uye zvekutarisa. Zvidzidzo zvekutevera uchishandisa mienzaniso mikuru zvinodiwa kusimbisa mhedzisiro yakawanikwa muchidzidzo ichi.
Pakati pezvidzidzo zvakaedza kururama kweSD mukufungidzira zera remazino, zvimwe zvakaratidza kururama kwakanyanya kupfuura ongororo yedu. Stepanovsky et al 35 vakashandisa mamodheru eSD makumi maviri nemaviri pamifananidzo yepanoramic yevagari vekuCzech 976 vane makore 2.7 kusvika 20.5 uye vakaedza kururama kwemodheru yega yega. Vakaongorora kukura kwemazino gumi nematanhatu epamusoro nepazasi kuruboshwe vachishandisa nzira yekupatsanura yakakurudzirwa naMoorrees et al 36. MAE inotangira pamakore 0.64 kusvika 0.94 uye RMSE inotangira pamakore 0.85 kusvika 1.27, ayo akarurama kupfuura mamodheru maviri eDM akashandiswa muchidzidzo ichi. Shen et al23 vakashandisa nzira yeCameriere kufungidzira zera remazino manomwe epermanent muruboshwe rwerudyi muvagari vekumabvazuva kweChina vane makore 5 kusvika 13 uye vakaienzanisa nemakore anofungidzirwa vachishandisa linear regression, SVM neRF. Vakaratidza kuti mamodheru ese matatu eDM ane kururama kwakanyanya zvichienzaniswa nefomura yechinyakare yeCameriere. MAE neRMSE muchidzidzo chaShen zvaive zvakaderera pane zviri muDM model muchidzidzo ichi. Kuwedzera kwekururama kwezvidzidzo zvakaitwa naStepanovsky et al. 35 naShen et al. 23 kungave kuri nekuda kwekuiswa kwevanhu vadiki mumienzaniso yavo yekudzidza. Nekuti fungidziro yezera yevatori vechikamu vane mazino ari kukura inova yakarurama sezvo huwandu hwemazino huchiwedzera panguva yekukura kwemazino, kururama kwenzira yekufungidzira zera inobuda kunogona kukanganiswa kana vatori vechikamu muchidzidzo vari vadiki. Pamusoro pezvo, kukanganisa kweMLP mukufungidzira zera kudiki zvishoma pane kweSLP, zvichireva kuti MLP yakarurama kupfuura SLP. MLP inoonekwa seyakanaka zvishoma pakufungidzira zera, pamwe nekuda kwezvikamu zvakavanzika muMLP38. Zvisinei, pane musiyano kune muenzaniso wekunze wevakadzi (SLP 1.45, MLP 1.49). Kuwanikwa kwekuti MLP yakarurama kupfuura SLP pakuongorora zera kunoda zvimwe zvidzidzo zvekuongorora.
Kushanda kweDM modhi uye nzira yechinyakare pakupatsanurana kwakaenzaniswawo. MaSD modhi ese akaedzwa uye nzira dzechinyakare paseti yebvunzo yemukati zvakaratidza mazinga ekusarura anogamuchirwa kune ane makore 18 ekuberekwa. Kunzwa kwevarume nevakadzi kwaive kwakakura kupfuura 87.7% uye 94.9%, zvichiteerana, uye kujeka kwaive kwakakura kupfuura 89.3% uye 84.7%. AUROC yemamodhi ese akaedzwa inodarikawo 0.925. Sekuziva kwedu, hapana chidzidzo chakaedza kushanda kweDM modhi ye18-year classification zvichibva pakukura kwemazino. Tinogona kuenzanisa mhedzisiro yechidzidzo ichi nekushanda kwekuisa muzvikamu kwemamodhi ekudzidza kwakadzama papanoramic radiographs. Guo et al.15 vakaverenga kushanda kweCNN-based deep learning modhi uye nzira yemaoko zvichibva panzira yaDemirjian yezera rakati. Kunzwa uye kujeka kwenzira yekushandisa nemaoko kwaive 87.7% uye 95.5%, zvichiteerana, uye kujeka uye kujeka kwemhando yeCNN kwakapfuura 89.2% uye 86.6%, zvichiteerana. Vakagumisa kuti mamodheru ekudzidza zvakadzama anogona kutsiva kana kupfuura ongororo yemaoko mukuisa muzvikamu zvezera. Mhedzisiro yechidzidzo ichi yakaratidza kushanda kwakafanana mukuisa muzvikamu; Zvinotendwa kuti kuisa muzvikamu uchishandisa mamodheru eDM kunogona kutsiva nzira dzechinyakare dzekuverenga zera. Pakati pemamodheru, DM LR ndiyo yaive modheru yakanakisa maererano nekunzwa kwemuenzaniso wevarume uye kujeka uye kujeka kwemuenzaniso wevakadzi. LR iri pechipiri mukuita kwevarume. Uyezve, LR inoonekwa seimwe yemamodheru eDM35 ari nyore kushandisa uye haina kuoma uye yakaoma kugadzirisa. Zvichibva pamhedzisiro iyi, LR yakaonekwa semodheru yakanakisa yekuisa muzvikamu kune vane makore gumi nemasere muvanhu vekuKorea.
Kazhinji, kururama kwekufungidzira zera kana kuita kwezvikamu paseti yebvunzo yekunze kwaive kwakashata kana kwakaderera zvichienzaniswa nemhedzisiro yeseti yebvunzo yemukati. Mimwe mishumo inoratidza kuti kururama kwezvikamu kana kushanda zvakanaka kunoderera kana fungidziro dzezera zvichibva pahuwandu hwevanhu vekuKorea dzichishandiswa kuvanhu vekuJapan5,39, uye maitiro akafanana akawanikwa muchidzidzo chino. Maitiro aya ekuderera akaonekwawo mumuenzaniso weDM. Saka, kuti tifungidzire zera nemazvo, kunyangwe pakushandisa DM mukuongorora, nzira dzinobva kudata revanhu veko, dzakadai senzira dzechinyakare, dzinofanira kusarudzwa5,39,40,41,42. Sezvo zvisiri pachena kana mamodheru ekudzidza zvakadzama anogona kuratidza maitiro akafanana, zvidzidzo zvinoenzanisa kururama kwezvikamu uye kushanda zvakanaka uchishandisa nzira dzechinyakare, mamodheru eDM, uye mamodheru ekudzidza zvakadzama pamienzaniso imwechete zvinodiwa kusimbisa kana huchenjeri hwekugadzira huchigona kukunda kusawirirana uku kwemarudzi mukuongorora kwezera diki.
Tinoratidza kuti nzira dzechinyakare dzinogona kutsiviwa nekufungidzira zera zvichibva pamuenzaniso weDM mukuita kwekuongorora zera re forensic muKorea. Takaonawo mukana wekushandisa kudzidza kwemuchina kwekuongorora zera re forensic. Zvisinei, pane zvipingamupinyi zvakajeka, zvakaita sehuwandu husina kukwana hwevatori vechikamu muchidzidzo ichi kuti vaone mhedzisiro yacho, uye kushaikwa kwezvidzidzo zvekare zvekuenzanisa nekusimbisa mhedzisiro yechidzidzo ichi. Mune ramangwana, zvidzidzo zveDM zvinofanirwa kuitwa nehuwandu hwakawanda hwemasampuli uye huwandu hwakasiyana hwevanhu kuti zvivandudze kushanda kwayo zvichienzaniswa nenzira dzechinyakare. Kuti zvionekwe kuti zvinoita here kushandisa njere dzekugadzira kufungidzira zera muhuwandu hwevanhu vakawanda, zvidzidzo zvemangwana zvinodiwa kuti zvienzanise kururama kwekuisa muzvikamu uye kushanda kweDM uye mamodheru ekudzidza kwakadzama nenzira dzechinyakare mumuenzaniso mumwe chete.
Chidzidzo ichi chakashandisa mifananidzo 2,657 yekunyora yakatorwa kubva kuvanhu vakuru vekuKorea nevekuJapan vane makore ari pakati pe15 ne23. MaX-ray ekuKorea akakamurwa kuita maX-ray ekudzidzisa mazana mapfumbamwe (makore 19.42 ± 2.65) uye maX-ray emukati mazana mapfumbamwe (makore 19.52 ± 2.59). Seti yekudzidziswa yakaunganidzwa panzvimbo imwe chete (Seoul St. Mary's Hospital), uye bvunzo yacho yakatorwa munzvimbo mbiri (Seoul National University Dental Hospital neYonsei University Dental Hospital). Takaunganidzawo maX-ray e857 kubva kune imwe data yakavakirwa pahuwandu hwevanhu (Iwate Medical University, Japan) yekuongororwa kwekunze. MaX-ray evaya vekuJapan (makore 19.31 ± 2.60) akasarudzwa seseti yebvunzo yekunze. Data rakaunganidzwa zvichitevera kufambira mberi kwekuongorora matanho ekukura kwemazino paX-ray dzepanoramic dzakatorwa panguva yekurapwa kwemazino. Data rese rakaunganidzwa rakanga risingazivikanwe kunze kwemurume kana mukadzi, zuva rekuzvarwa uye zuva reX-ray. Zvinodiwa zvekubatanidza uye zvekubvisa zvaive zvakafanana nezvidzidzo zvakaburitswa kare 4, 5. Zera chairo remuenzaniso rakaverengerwa nekubvisa zuva rekuzvarwa kubva pazuva rakatorwa radiograph. Boka remuenzaniso rakakamurwa kuita mapoka mapfumbamwe ezera. Kugoverwa kwezera nebonde kunoratidzwa muTafura 3 Chidzidzo ichi chakaitwa zvichienderana neChiziviso cheHelsinki uye chakatenderwa neInstitutional Review Board (IRB) yeSeoul St. Mary's Hospital yeCatholic University of Korea (KC22WISI0328). Nekuda kwekugadzirwa kwechidzidzo ichi, mvumo yeruzivo yaisagona kuwanikwa kubva kuvarwere vese vaiongororwa radiographic kuti varapwe. Seoul Korea University St. Mary's Hospital (IRB) yakabvisa chinodiwa chekubvumidzwa kweruzivo.
Matanho ekukura emazino echipiri neechitatu e-bimaxillary akaongororwa zvichienderana neDemircan criteria25. Zino rimwe chete rakasarudzwa kana rudzi rumwe rwezino rwakawanikwa kuruboshwe nerudyi rweshaya yega yega. Kana mazino akafanana kumativi ese ari pamatanho akasiyana ekukura, zino rine danho rekukura rakaderera rakasarudzwa kuti riverenge kusava nechokwadi kwezera rakafungidzirwa. MaX-raygraph zana akasarudzwa zvisina kurongeka kubva kuboka rekudzidziswa akapihwa nevanoongorora vaviri vane ruzivo kuti vaedze kuvimbika kwevanoongorora mushure mekuongororwa kwekutanga kuti vaone danho rekukura kwemazino. Kuvimbika kwevanoongorora mukati memaziso kwakaongororwa kaviri mumwedzi mitatu nemuongorori wekutanga.
Danho rebonde nekukura kwemolars yechipiri neyechitatu yeshaya yega yega muchikamu chekudzidzisa zvakayerwa nemunhu anoona nezvekutanga akadzidziswa nemamodheru akasiyana eDM, uye zera chairo rakaiswa sechinangwa chekukosha. Mamodheru eSLP neMLP, anoshandiswa zvakanyanya mukudzidza kwemuchina, akaedzwa achipesana nemaalgorithms eregression. Modheru yeDM inosanganisa mabasa elinear uchishandisa matanho ekukura emazino mana uye inosanganisa data iri kufungidzira zera. SLP ndiyo network iri nyore yeneural uye haina ma layers akavanzwa. SLP inoshanda zvichibva pakutumira kwe threshold pakati pema nodes. Modheru yeSLP muregression yakafanana nemasvomhu ne multiple linear regression. Kusiyana neSLP modhi, modhi yeMLP ine ma layers akavanzwa akawanda ane mabasa e nonlinear activation. Kuedza kwedu kwakashandisa layer yakavanzwa ine ma nodes makumi maviri chete akavanzwa ane mabasa e nonlinear activation. Shandisa gradient descent senzira ye optimization uye MAE neRMSE sebasa rekurasikirwa kudzidzisa modhi yedu yekudzidza kwemuchina. Modhi yeregression yakawanikwa yakanakisa yakashandiswa kuma test sets emukati nekunze uye zera remazino rakayerwa.
Algorithm yekupatsanura yakagadzirwa inoshandisa kukura kwemazino mana paseti yekudzidzira kufanotaura kana sampuro ine makore gumi nemasere kana kwete. Kuti tigadzire modhi iyi, takawana maalgorithms manomwe ekudzidza kwemuchina anomiririra6,43: (1) LR, (2) KNN, (3) SVM, (4) DT, (5) RF, (6) XGBoost, uye (7) MLP. LR ndeimwe yemaalgorithms ekupatsanura anoshandiswa zvakanyanya44. Ialgorithm yekudzidza inotariswa inoshandisa regression kufanotaura mukana wekuti data rive rechikamu chakati kubva pa0 kusvika pa1 uye inoisa data muchikamu chine mukana wakawanda zvichibva pane mukana uyu; inonyanya kushandiswa pakupatsanura binary. KNN ndeimwe yemaalgorithms ekupinza emuchina ari nyore45. Kana ikapihwa data idzva rekupinda, inowana data rek riri pedyo neseti iripo uye yozoiisa mukirasi ine frequency yepamusoro. Tinoisa 3 yenhamba yevavakidzani vanofungwa nezvayo (k). SVM ialgorithm inowedzera daro riri pakati pemakirasi maviri nekushandisa kernel function kuwedzera nzvimbo yemutsara munzvimbo isina mutsara inonzi fields46. Pamuenzaniso uyu, tinoshandisa bias = 1, simba = 1, uye gamma = 1 se hyperparameters ye polynomial kernel. DT yakashandiswa munzvimbo dzakasiyana-siyana se algorithm yekuparadzanisa data rese mumapoka akati wandei nekumiririra mitemo yesarudzo muchimiro chemuti47. Muenzaniso uyu wakagadzirwa nenhamba shoma yezvinyorwa pa node imwe neimwe ye2 uye unoshandisa Gini index sechiyero chemhando. RF inzira yekubatanidza inosanganisa maDT akawanda kuti ivandudze mashandiro uchishandisa nzira ye bootstrap aggregation inogadzira weak classifier yemuenzaniso wega wega nekudhirowa samples dzehukuru hwakafanana kakawanda kubva kudata rekutanga48. Takashandisa miti zana, kudzika kwemiti gumi, saizi imwe chete ye node, uye Gini admixture index senzira yekuparadzanisa node. Kupatsanurwa kwedata idzva kunosarudzwa nevhoti yeruzhinji. XGBoost i algorithm inosanganisa matekiniki ekusimudzira ichishandisa nzira inotora sekudzidzisa data kukanganisa pakati pezvakakosha chaizvo uye zvakafanotaurwa zvemuenzaniso wekare uye kuwedzera kukanganisa uchishandisa gradients49. Iyi ialgorithm inoshandiswa zvakanyanya nekuda kwekushanda kwayo zvakanaka uye kushanda zvakanaka kwezviwanikwa, pamwe nekuvimbika kwakanyanya sebasa rekugadzirisa overfitting. Modhi iyi ine mavhiri ekutsigira mazana mana. MLP inetwork yetsinga umo perceptron imwe kana kupfuura dzinogadzira ma layer akawanda ne layer imwe kana kupfuura yakavanzika pakati pe input ne output layers38. Uchishandisa izvi, unogona kuita non-linear classification apo paunowedzera input layer wowana result value, fungidziro yemhedzisiro inoenzaniswa nemhedzisiro chaiyo uye error inopararira kumashure. Takagadzira layer yakavanzwa ine 20 hidden neurons mu layer yega yega. Modhi yega yega yatakagadzira yakashandiswa kumaseti emukati nekunze kuyedza classification performance nekuverenga sensitivity, specificity, PPV, NPV, uye AUROC. Sensitivity inotsanangurwa se ratio yesample inofungidzirwa kuva nemakore gumi nemasere ekuberekwa kana kupfuura kune sample inofungidzirwa kuva nemakore gumi nemasere ekuberekwa kana kupfuura. Specificity i part of samples under 18 years ekuberekwa uye avo vanofungidzirwa kuva pasi pemakore gumi nemasere ekuberekwa.
Matanho emazino akaongororwa muchikamu chekudzidzisa akachinjwa kuita matanho ekuverenga nhamba kuti aongorore nhamba. Kudzokorora kwakawanda kwemutsara uye kwelogistic kwakaitwa kuti pave nemamodheru ekufanotaura ebonde rega rega uye kuwana maforomaru ekudzoka anogona kushandiswa kufungidzira zera. Takashandisa maforomaru aya kufungidzira zera remazino kumaseti ese ebvunzo emukati nekunze. Tafura 4 inoratidza mamodheru ekudzoka uye ekuisa muzvikamu akashandiswa muchidzidzo ichi.
Kuvimbika kwemukati nemukati memuoni kwakaverengerwa tichishandisa nhamba yaCohen yekappa. Kuti tiedze kururama kweDM uye mamodheru ekudzokorora echinyakare, takaverenga MAE neRMSE tichishandisa makore anofungidzirwa uye chaiwo emaseti ekuyedza emukati nekunze. Zvikanganiso izvi zvinowanzo shandiswa kuongorora kururama kwekufanotaura kwemuenzaniso. Kana kukanganisa kudiki, kururama kwekufanotaura kwacho kwakakwira24. Enzanisa MAE neRMSE yemaseti ekuyedza emukati nekunze akaverengerwa uchishandisa DM uye kudzoreredzwa kwechinyakare. Kushanda kwechikamu chegumi nemasere che cutoff muhuwandu hwechinyakare kwakaongororwa uchishandisa tafura ye2 × 2 contingency. Kunzwisisika kwakaverengerwa, specificity, PPV, NPV, uye AUROC yeseti yekuyedza kwakaenzaniswa nemitengo yakayerwa yemuenzaniso wekuenzanisa weDM. Data rinoratidzwa seavhareji ± standard deviation kana nhamba (%) zvichienderana nehunhu hwedata. Two-sided P values <0.05 dzakaonekwa sedzakakosha muhuwandu. Ongororo dzese dzehuwandu hwehuwandu dzakaitwa uchishandisa SAS version 9.4 (SAS Institute, Cary, NC). Modhi yeDM regression yakashandiswa muPython uchishandisa Keras50 2.2.4 backend uye Tensorflow51 1.8.0 kunyanya pakushanda kwemasvomhu. Modhi yeDM classification yakashandiswa muWaikato Knowledge Analysis Environment uye Konstanz Information Miner (KNIME) 4.6.152 analysis platform.
Vanyori vanobvuma kuti data rinotsigira mhedziso yechidzidzo rinogona kuwanikwa munyaya yacho uye mune zvimwe zvinyorwa. Datasets dzakagadzirwa uye/kana dzakaongororwa panguva yechidzidzo dzinowanikwa kubva kumunyori anoenderana kana zvikakumbirwa zvine musoro.
Ritz-Timme, S. nevamwe. Kuongororwa kwezera: hunyanzvi hwepamusoro hwekusangana nezvinodiwa chaizvo zvekudzidzira kwezvekuferefeta. nyika dzese. J. Mishonga yepamutemo. 113, 129–136 (2000).
Schmeling, A., Reisinger, W., Geserik, G., naOlze, A. Mamiriro aripo pari zvino ekuongororwa kwezera revanhu vari kurarama nekuda kwezvikonzero zvekupomerwa mhosva. Forensics. mushonga. Pathology. 1, 239–246 (2005).
Pan, J. nevamwe. Nzira yakagadziriswa yekuongorora zera remazino revana vane makore mashanu kusvika gumi nematanhatu kumabvazuva kweChina. clinical. Oral survey. 25, 3463–3474 (2021).
Lee, SS nezvimwewo. Kurongeka kwekukura kwemazino emusana wechipiri newechitatu muKorea uye kushandiswa kwawo pakuongorora zera remusana. nyika dzese. J. Legal medicine. 124, 659–665 (2010).
Oh, S., Kumagai, A., Kim, SY naLee, SS Kururama kwekufungidzira zera uye kufungidzira kwechikamu chemakore gumi nemasere zvichibva pakukura kwemolars yechipiri neyechitatu muKorea neJapan. PLoS ONE 17, e0271247 (2022).
Kim, JY, nevamwe. Kuongororwa kwedata kwakavakirwa pakudzidza kwemuchina usati wavhiyiwa kunogona kufanotaura mhedzisiro yekurapwa kwekuvhiyiwa kwekurara muvarwere vane OSA. sainzi. Mushumo 11, 14911 (2021).
Han, M. nevamwe. Kufungidzira zera kwakarurama kubva mukudzidza kwemuchina nekupindira kwevanhu kana kuti pasina? nyika dzese. J. Legal medicine. 136, 821–831 (2022).
Khan, S. naShaheen, M. Kubva paKuchera Data kusvika paKuchera Data. J.Information. the science. https://doi.org/10.1177/01655515211030872 (2021).
Khan, S. naShaheen, M. WisRule: Nzira Yekutanga Yekuziva Yekushandisa Mutemo weKusanganisa Kugadzira Migodhi. J.Information. sainzi. https://doi.org/10.1177/01655515221108695 (2022).
Shaheen M. naAbdullah U. Karm: Kuchera data kwechinyakare zvichibva pamitemo yekubatanidza yakavakirwa pamamiriro ezvinhu. kuverenga. Matt. enderera mberi. 68, 3305–3322 (2021).
Muhammad M., Rehman Z., Shaheen M., Khan M. naHabib M. Kuona kufanana kwemazwi kwakavakirwa pakudzidza kwakadzama uchishandisa data rechinyorwa. inform. technologies. control. https://doi.org/10.5755/j01.itc.49.4.27118 (2020).
Tabish, M., Tanoli, Z., naShahin, M. Sisitimu yekuziva zviitiko mumavhidhiyo emitambo. multimedia. Zvishandiso zveMashandisirwo https://doi.org/10.1007/s11042-021-10519-6 (2021).
Halabi, SS nevamwe. RSNA Machine Learning Challenge muPediatric Bone Age. Radiology 290, 498–503 (2019).
Li, Y. nevamwe. Kuongororwa kwezera remudumbu kubva paX-rays dzechiuno uchishandisa kudzidza kwakadzama. EURO. radiation. 29, 2322–2329 (2019).
Guo, YC, nevamwe. Kurongeka kwezera kwakarurama uchishandisa nzira dzemaoko uye network dzakadzika dze convolutional neural kubva kumifananidzo ye ortografia. internationality. J. Legal medicine. 135, 1589–1597 (2021).
Alabama Dalora nevamwe. Kuongororwa kwezera remapfupa vachishandisa nzira dzakasiyana dzekudzidza kwemuchina: ongororo yemabhuku yakarongeka uye ongororo yemeta. PLoS ONE 14, e0220242 (2019).
Du, H., Li, G., Cheng, K., naYang, J. Kufungidzirwa kwezera revanhu vemuAfrica nevekuChina zvichienderana nehuwandu hwemamota ekutanga emudumbu uchishandisa cone-beam computed tomography. internationality. J. Legal medicine. 136, 811–819 (2022).
Kim S., Lee YH, Noh YK, Park FK naOh KS Kuona mapoka ezera revanhu vapenyu vachishandisa mifananidzo yehungwaru hwekugadzira yemolars dzekutanga. sainzi. Mushumo 11, 1073 (2021).
Stern, D., Payer, C., Giuliani, N., naUrschler, M. Kuongororwa kwezera otomatiki uye kupatsanurwa kwezera revanhu vazhinji kubva kudata reMRI rine variate. IEEE J. Biomed. Zviziviso zveHutano. 23, 1392–1403 (2019).
Cheng, Q., Ge, Z., Du, H. naLi, G. Kufungidzira zera kwakavakirwa pakupatsanurwa kwe3D pulp chamber ye first molars kubva ku cone beam computed tomography nekubatanidza kudzidza kwakadzama uye mazinga emazinga. internationality. J. Legal medicine. 135, 365–373 (2021).
Wu, WT, nevamwe. Kuchera data mu data guru rekiriniki: madhatabhesi akajairika, matanho, uye nzira dzekushandisa. World. medicine. resource. 8, 44 (2021).
Yang, J. nevamwe. Nhanganyaya kuMedical Databases neData Mining Technologies munguva yeBig Data. J. Avid. Basic medicine. 13, 57–69 (2020).
Shen, S. nevamwe. Nzira yaCamerer yekuongorora zera remazino uchishandisa machine learning. BMC Oral Health 21, 641 (2021).
Galliburg A. nevamwe. Kuenzanisa nzira dzakasiyana dzekudzidza kwemuchina dzekufanotaura zera remazino uchishandisa nzira yeDemirdjian staging. internationality. J. Legal medicine. 135, 665–675 (2021).
Demirdjian, A., Goldstein, H. naTanner, JM Sisitimu itsva yekuongorora zera remazino. snort. biology. 45, 211–227 (1973).
Landis, JR, naKoch, GG. Zviyero zvekubvumirana kwevacherechedzi padata rezvikamu. Biometrics 33, 159–174 (1977).
Bhattacharjee S, Prakash D, Kim C, Kim HK naChoi HK. Kuongororwa kwemavara, morphological uye statistical kwe two-dimensional magnetic resonance imaging uchishandisa artificial intelligence techniques yekusiyanisa mabundu ekutanga europi. Ruzivo rwehutano. resource. https://doi.org/10.4258/hir.2022.28.1.46 (2022).
Nguva yekutumira: Ndira-04-2024