T\u0131p alan\u0131nda tan\u0131 s\u00fcre\u00e7lerini k\u00f6kten d\u00f6n\u00fc\u015ft\u00fcrme potansiyeline sahip ileri bir yapay zeka modeli, a\u011f\u0131z i\u00e7i epitel displazisinin tespit ve derecelendirilmesinde \u00e7\u0131\u011f\u0131r a\u00e7an bir geli\u015fme olarak g\u00fcndeme geldi. A\u011f\u0131z epitel displazisi (OED), a\u011f\u0131z mukozas\u0131nda h\u00fccresel davran\u0131\u015flar\u0131n anormal hale gelmesiyle karakterize edilen ve oral kanser ilerleyi\u015finde kritik \u00f6neme sahip precancer\u00f6z bir durumdur. Bu hastal\u0131\u011f\u0131n do\u011fru ve hassas bir \u015fekilde derecelendirilmesi, uygun ve zaman\u0131nda m\u00fcdahaleyi m\u00fcmk\u00fcn k\u0131larken; patologlar\u0131n uzmanl\u0131\u011f\u0131na dayanan histopatolojik incelemeler ise olduk\u00e7a karma\u015f\u0131k ve subjektif de\u011ferlendirmelere sahiptir. Tahran T\u0131p Bilimleri \u00dcniversitesinden ara\u015ft\u0131rma ekibi, bu zorluklara \u00e7\u00f6z\u00fcm getirmek ve tan\u0131 do\u011frulu\u011funu art\u0131rmak amac\u0131yla Vision Transformer (ViT) mimarisi tabanl\u0131 yeni bir yapay zeka sistemi geli\u015ftirdi.<\/p>\n
Histopatoloji alan\u0131ndaki tan\u0131sal s\u00fcre\u00e7ler, boyanm\u0131\u015f doku kesitlerinin mikroskobik incelemesine dayan\u0131r ve bu, deneyimli patoloji uzmanlar\u0131n\u0131n uzmanl\u0131\u011f\u0131n\u0131 gerektiren zor ve yorucu bir i\u015ftir. Ayr\u0131ca, g\u00f6zlemciye g\u00f6re de\u011ferlendirmelerde \u00f6nemli farkl\u0131l\u0131klar olabilmektedir. Geli\u015ftirilen yapay zeka sistemi, bu analitik i\u015f ak\u0131\u015f\u0131n\u0131 otomatikle\u015ftirerek daha kesin tan\u0131lar sa\u011flamay\u0131, insan kaynakl\u0131 hatalar\u0131 minimize etmeyi ve maliyetleri azaltmay\u0131 hedeflemektedir. \u00c7al\u0131\u015fmada, kurumlar\u0131n ar\u015fivlerinden toplanan 218 histopatolojik slayt g\u00f6r\u00fcnt\u00fcs\u00fcne ek olarak, a\u00e7\u0131k eri\u015fimli veri tabanlar\u0131ndan temin edilen g\u00f6r\u00fcnt\u00fcler kullan\u0131ld\u0131. \u0130ki ba\u011f\u0131ms\u0131z a\u011f\u0131z patolo\u011fu taraf\u0131ndan, 2022 D\u00fcnya Sa\u011fl\u0131k \u00d6rg\u00fct\u00fc (WHO) derecelendirme sistemi do\u011frultusunda; hafif, orta ve a\u011f\u0131r displazi, d\u00fc\u015f\u00fck risk, y\u00fcksek risk olmak \u00fczere ikili s\u0131n\u0131fland\u0131rma ve ayr\u0131ca normal dokuyu i\u00e7eren kategoriler titizlikle etiketlendi.<\/p>\n
Bu yenilik\u00e7i ba\u015far\u0131n\u0131n teknik oda\u011f\u0131, geleneksel konvol\u00fcsyonel sinir a\u011flar\u0131ndan (CNN) farkl\u0131 olarak, self-attention mekanizmalar\u0131n\u0131 kullanarak t\u00fcm g\u00f6r\u00fcnt\u00fcdeki uzun menzilli ili\u015fkileri yakalayan Vision Transformer algoritmas\u0131d\u0131r. CNN\u2019lerin sadece lokal \u00f6zellikleri \u00e7\u0131karabildi\u011fi, oysa Transformer\u2019lar\u0131n daha geni\u015f bir ba\u011flamsal anlay\u0131\u015fla histolojik yap\u0131lar\u0131 analiz edebildi\u011fi bu yakla\u015f\u0131m, OED g\u00f6r\u00fcnt\u00fclerinin karma\u015f\u0131k yap\u0131s\u0131n\u0131 daha iyi kavray\u0131p s\u0131n\u0131fland\u0131rmada \u00fcst\u00fcnl\u00fck sa\u011flamaktad\u0131r. Ara\u015ft\u0131rmac\u0131lar, ViT modelinin performans\u0131n\u0131 VGG16 ve \u00f6zel tasarlanm\u0131\u015f bir ConvNet modeli ile kar\u015f\u0131la\u015ft\u0131rarak de\u011ferlendirdi.<\/p>\n
Veri \u00f6n i\u015flemi a\u015famas\u0131, ham histopatolojik slaytlar\u0131n lokal doku b\u00f6lgelerini temsil eden birbirinden ba\u011f\u0131ms\u0131z \u2018yama\u2019 (patch) halinde par\u00e7alara ayr\u0131lmas\u0131yla ger\u00e7ekle\u015fti. Elde edilen 2.545 d\u00fc\u015f\u00fck risk, 2.054 y\u00fcksek risk yamas\u0131; bunlar\u0131n i\u00e7inde 726 hafif, 831 orta ve 449 a\u011f\u0131r displazi yamas\u0131 ile 937 normal doku yamas\u0131, detayl\u0131 s\u0131n\u0131fland\u0131rma ve model e\u011fitimi i\u00e7in rich bir veri seti olu\u015fturdu. Bu y\u00fcksek \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc ayr\u0131nt\u0131l\u0131 yap\u0131, modelin genelleme kabiliyetini art\u0131r\u0131rken, ger\u00e7ek klinik ortamlar i\u00e7in i\u00e7 ge\u00e7erli\u011fi destekleyen sa\u011flam bir e\u011fitim zemini sa\u011flad\u0131.<\/p>\n
Say\u0131sal de\u011ferlendirmeler, Vision Transformer\u2019\u0131n \u00fcst\u00fcn tahmin kapasitesini net bir \u015fekilde ortaya koydu. WHO\u2019nun \u00fc\u00e7 s\u0131n\u0131fl\u0131 grading sistemine g\u00f6re ViT modelinin %94 do\u011fruluk elde etmesi, VGG16 ve ConvNet modellerinin s\u0131ras\u0131yla %86 ve %88\u2019lik ba\u015far\u0131 oranlar\u0131n\u0131 \u00f6nemli \u00f6l\u00e7\u00fcde ge\u00e7ti. D\u00f6rt s\u0131n\u0131f\u0131n yer ald\u0131\u011f\u0131, ikili risk s\u0131n\u0131fland\u0131rmas\u0131 ile normal doku ayr\u0131m\u0131n\u0131 i\u00e7eren zorlu senaryoda ViT\u2019in ba\u015far\u0131 oran\u0131 %97\u2019ye ula\u015f\u0131rken, VGG16 %79 ve ConvNet %88\u2019de kald\u0131. Bu farklar, ViT\u2019nin displazi derecelerinin karma\u015f\u0131k morfolojik varyasyonlar\u0131n\u0131 ay\u0131rt etme konusundaki \u00fcst\u00fcn yetkinli\u011fini ortaya koydu.<\/p>\n
Performans g\u00f6stergelerinin \u00f6tesinde, bu \u00e7al\u0131\u015fma Vision Transformer\u2019lar\u0131n patoloji pratiklerine getirece\u011fi yap\u0131sal ve i\u015flevsel yeniliklere de \u0131\u015f\u0131k tuttu. ViT\u2019nin k\u00fcresel mek\u00e2nsal ili\u015fkileri modelleme becerisi, h\u00fccresel farkl\u0131la\u015fma, doku mimarisi bozukluklar\u0131 ve stromal de\u011fi\u015fikliklerin incelikli s\u0131n\u0131fland\u0131r\u0131lmas\u0131n\u0131 m\u00fcmk\u00fcn k\u0131ld\u0131. Bu, AI\u2019n\u0131n geleneksel algoritmalar\u0131n \u00f6tesinde, sadece yard\u0131mc\u0131 de\u011fil, ayn\u0131 zamanda patolo\u011fun de\u011ferlendirmesini g\u00fc\u00e7lendiren bir ara\u00e7 olabilece\u011finin kan\u0131t\u0131 oldu.<\/p>\n
Ara\u015ft\u0131rman\u0131n \u00f6nemi sadece a\u011f\u0131z patolojisi alan\u0131 ile s\u0131n\u0131rl\u0131 kalmay\u0131p, d\u00fcnya genelinde s\u0131k g\u00f6r\u00fclen oral kanserlerin erken te\u015fhisinde hayati bir ad\u0131m olarak de\u011ferlendiriliyor. Erken tan\u0131 sayesinde, tedavi s\u00fcre\u00e7leri iyile\u015fmekte ve hastal\u0131k y\u00fck\u00fc bir\u00e7ok \u00fclkede azalabilmektedir. Dijital patoloji ile birlikte bu yapay zeka tabanl\u0131 tan\u0131 sistemi, uzman eri\u015fiminin k\u0131s\u0131tl\u0131 oldu\u011fu b\u00f6lgelerde bile tekrarlanabilir, h\u0131zl\u0131 ve nesnel sonu\u00e7lar sunarak sa\u011fl\u0131k hizmetlerini d\u00f6n\u00fc\u015ft\u00fcrmeye aday.<\/p>\n
Ayr\u0131ca, bu \u00e7al\u0131\u015fma do\u011fal dil i\u015fleme ve bilgisayar g\u00f6r\u00fcs\u00fc gibi alanlarda geli\u015ftirilen son teknoloji AI modellerinin t\u0131bbi uygulamalara ba\u015far\u0131yla adapte edilmesinde bir d\u00f6n\u00fcm noktas\u0131 olu\u015fturdu. G\u00f6r\u00fcnt\u00fc tan\u0131ma alan\u0131nda \u00e7\u0131\u011f\u0131r a\u00e7an Vision Transformer\u2019lar\u0131n heterojen ve karma\u015f\u0131k t\u0131bbi g\u00f6r\u00fcnt\u00fcler ile ba\u015fa \u00e7\u0131kmadaki ba\u015far\u0131s\u0131, biyomedikal teknolojide yeni ara\u015ft\u0131rma kap\u0131lar\u0131n\u0131 aral\u0131yor. OED derecelendirmesindeki bu uygulama, patoloji uzmanl\u0131\u011f\u0131 ile yapay zekan\u0131n g\u00fc\u00e7l\u00fc bir b\u00fct\u00fcnle\u015fmesini simgeliyor.<\/p>\n
Bununla birlikte, ara\u015ft\u0131rmac\u0131lar klinik uygulamalar \u00f6ncesinde baz\u0131 zorluklar\u0131n \u00fcstesinden gelinmesi gerekti\u011fini de vurguluyor. Daha geni\u015f ve \u00e7ok merkezli veri k\u00fcmelerinde do\u011frulama yap\u0131lmas\u0131, farkl\u0131 boyama protokollerinin entegrasyonu ve standartla\u015ft\u0131r\u0131lm\u0131\u015f yapay zeka da\u011f\u0131t\u0131m altyap\u0131lar\u0131n\u0131n olu\u015fturulmas\u0131 en \u00f6nemli \u00f6ncelikler aras\u0131nda. Ayr\u0131ca, AI kararlar\u0131n\u0131n \u015feffaf ve a\u00e7\u0131klanabilir bi\u00e7imde sunulmas\u0131, klinisyenlerin g\u00fcvenini kazanmak ve etik uyum sa\u011flamak bak\u0131m\u0131ndan zorunlu g\u00f6r\u00fcl\u00fcyor.<\/p>\n
Sonu\u00e7 olarak, bu \u00e7al\u0131\u015fma AI destekli histopatolojide yeni bir d\u00f6neme i\u015faret ediyor. Neredeyse insan seviyesindeki do\u011fruluk ve karma\u015f\u0131k histolojik yap\u0131lar\u0131 anlama kapasitesiyle Vision Transformer tabanl\u0131 model, ba\u011f\u0131ms\u0131z ya da tamamlay\u0131c\u0131 bir tan\u0131 arac\u0131 olarak i\u015flev g\u00f6rebilir. Tan\u0131 s\u00fcre\u00e7lerini h\u0131zland\u0131rd\u0131\u011f\u0131 gibi, kaliteli tan\u0131y\u0131 evrenselle\u015ftirerek \u00f6zellikle kaynaklar\u0131n s\u0131n\u0131rl\u0131 oldu\u011fu b\u00f6lgelerde hastal\u0131k y\u00fck\u00fcn\u00fcn azalt\u0131lmas\u0131na destek olabilir.<\/p>\n
Bu ara\u015ft\u0131rma AI yeniliklerinin t\u0131bbi g\u00f6r\u00fcnt\u00fcleme analizindeki geleneksel yetersizlikleri a\u015fmada ne kadar etkili olabilece\u011fini g\u00f6stermektedir. Vision Transformer modeli, sa\u011fl\u0131k profesyonellerine kapsaml\u0131 diagnostik yetenekler sunarak oral patolojide hassasiyet temelli bir paradigman\u0131n \u00f6nc\u00fcs\u00fc oldu. S\u00fcregelen geli\u015fmeler ve disiplinleraras\u0131 i\u015f birlikleri, yapay zeka destekli patolojinin klinik ak\u0131\u015fa kesintisiz entegrasyonunu, erken tan\u0131y\u0131 ve hastalar\u0131n ya\u015fam kalitesinin y\u00fckseltilmesini sa\u011flayacak.<\/p>\n
Ayr\u0131ca, Vision Transformer\u2019lar\u0131n ba\u015far\u0131l\u0131 kullan\u0131m\u0131, a\u011f\u0131z epitel displazisinin \u00f6tesinde di\u011fer karma\u015f\u0131k histopatolojik tan\u0131 alanlar\u0131nda da benzer teknolojilerin denenmesini te\u015fvik ediyor. Kanser karakterizasyonu, prognozland\u0131rma ve ki\u015fiselle\u015ftirilmi\u015f tedavi s\u00fcre\u00e7lerinde molek\u00fcler ve genomik verilerle b\u00fct\u00fcnle\u015ftirilen yapay zeka yakla\u015f\u0131mlar\u0131, gelecekte tan\u0131 algoritmalar\u0131n\u0131 yeniden \u015fekillendirebilir.<\/p>\n
\u00d6zetle, yapay zeka ve patolojinin kesi\u015fim noktas\u0131nda yeni bir \u00e7a\u011f ba\u015flayabilir. Vision Transformer\u2019lar\u0131n mikroskobik doku morfolojisini \u00e7\u00f6z\u00fcmlemedeki potansiyeli, sa\u011fl\u0131k hizmetlerinde yapay zekinin geni\u015f perspektifte kullan\u0131labilece\u011finin somut \u00f6rne\u011fi oldu. Bu \u00e7al\u0131\u015fma, hesaplama bilimindeki son geli\u015fmeler ile klinik uzmanl\u0131\u011f\u0131n birle\u015ferek daha h\u0131zl\u0131, ak\u0131ll\u0131 ve eri\u015filebilir tan\u0131 ara\u00e7lar\u0131na giden yolu a\u00e7t\u0131\u011f\u0131n\u0131 g\u00f6steriyor.<\/p>\n
Ara\u015ft\u0131rma Konusu<\/strong>: Artificial intelligence application in grading histopathological images of oral epithelial dysplasia using Vision Transformer deep learning algorithms.<\/p>\n Makale Ba\u015fl\u0131\u011f\u0131<\/strong>: Artificial intelligence based vision transformer application for grading histopathological images of oral epithelial dysplasia: a step towards AI-driven diagnosis.<\/p>\n Web References<\/strong>: https:\/\/doi.org\/10.1186\/s12885-025-14193-x<\/p>\n Doi Referans<\/strong>: https:\/\/doi.org\/10.1186\/s12885-025-14193-x<\/p>\n Resim Credits<\/strong>: Scienmag.com<\/p>\n Anahtar Kelimeler<\/strong>: ileri AI modelleri sa\u011fl\u0131kta, yapay zeka t\u0131bbi tan\u0131da, AI performans\u0131 doku g\u00f6r\u00fcnt\u00fc analizinde, otomatik tan\u0131 sistemleri, histopatolojik inceleme otomasyonu, oral patolojide do\u011frulu\u011fun art\u0131r\u0131lmas\u0131, patoloji de\u011ferlendirme varyabilitesi, kanser tespiti i\u00e7in makine \u00f6\u011frenimi, a\u011f\u0131z kanseri \u00f6nc\u00fcs\u00fc tan\u0131s\u0131, a\u011f\u0131z epitel displazisi tespiti, Tahran \u00dcniversitesi AI ara\u015ft\u0131rmalar\u0131, Vision Transformer patolojide<\/p>\n","protected":false},"excerpt":{"rendered":" T\u0131p alan\u0131nda tan\u0131 s\u00fcre\u00e7lerini k\u00f6kten d\u00f6n\u00fc\u015ft\u00fcrme potansiyeline sahip ileri bir yapay zeka modeli, a\u011f\u0131z i\u00e7i epitel displazisinin tespit ve derecelendirilmesinde \u00e7\u0131\u011f\u0131r a\u00e7an bir geli\u015fme olarak g\u00fcndeme geldi. A\u011f\u0131z epitel displazisi (OED), a\u011f\u0131z mukozas\u0131nda h\u00fccresel davran\u0131\u015flar\u0131n anormal hale gelmesiyle karakterize edilen ve oral kanser ilerleyi\u015finde kritik \u00f6neme sahip precancer\u00f6z bir durumdur. Bu hastal\u0131\u011f\u0131n do\u011fru ve hassas…<\/p>\n","protected":false},"author":1,"featured_media":3932,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_yoast_wpseo_title":"","_yoast_wpseo_metadesc":"","_yoast_wpseo_focuskw":"","rank_math_title":"","rank_math_description":"","rank_math_focus_keyword":"","_wpan_schema_json_ld":"","_wpan_ai_seo_metadata":"","_wpan_ai_seo_status":"","_wpan_ai_seo_policy":"","_wpan_ai_seo_faq_block":"","_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[28],"tags":[3358,3360,3359,3357,3356],"tmauthors":[],"class_list":{"0":"post-3931","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-kanser","8":"tag-agiziciepiteldisplazisitespiti","9":"tag-histopatolojikslaytgoruntuleme","10":"tag-oralkanseroncesitaniyontemleri","11":"tag-visiontransformerhistopatolojianalizi","12":"tag-yapayzekaoraldisplazitanisi"},"jetpack_featured_media_url":"https:\/\/haber360.com\/wp-content\/uploads\/2025\/04\/Yapay-Zeka-Vision-Transformer-ile-Oral-Displazi-Tanisi-1745618369.jpg","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/posts\/3931","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/comments?post=3931"}],"version-history":[{"count":0,"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/posts\/3931\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/media\/3932"}],"wp:attachment":[{"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/media?parent=3931"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/categories?post=3931"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/tags?post=3931"},{"taxonomy":"tmauthors","embeddable":true,"href":"https:\/\/haber360.com\/index.php\/wp-json\/wp\/v2\/tmauthors?post=3931"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}