BITEWING AĞIZ İÇİ RADYOGRAFİK GÖRÜNTÜLERDE DERİN ÖĞRENME İLE DİŞ SEGMENTASYONU
Yükleniyor...
Dosyalar
Tarih
2021-10
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Bu çalışmada, diş sağlığı alanında detaylı incelenmesi gereken dişlerin tedavi ve tanı sürecinde kullanılan bitewing (ısırma kanatlı) ağız içi radyografi görüntüleri üzerinde dişlerin segmentasyonu sağlanmıştır. Bitewing görüntüleme yöntemi, ağız içerisinde doğrudan görülemeyen, özellikle küçük ve büyük azı dişlerinin ara yüzlerindeki çürüklerin ve restorasyonların altında tekrarlayan çürüklerin tespiti için ideal bir tanı yöntemidir. Radyolojik görüntülerin uzmanlar tarafından raporlanması el ile yapılan, zaman alan ve dikkat gerektiren işlemlerdir. Bu işlemlerin başarılı ve hızlı bir şekilde gerçekleştirilebilmesi için derin öğrenme tabanlı bir yöntemle desteklenmesi amaçlanmıştır. Çalışma kapsamında kullanılan bitewing diş görüntüleriyle ilgili yasal izinler alınmıştır ve çalışmada kullanılan veri kümesi, Ordu Diş Hekimliği Fakültesi Ağız, Diş ve Çene Radyolojisi Anabilim Dalı radyoloji görüntü arşivinden elde edilmiştir. Bu çalışmada, Evrişimli Sinir Ağları (CNN) türünden olan Bölgesel Tabanlı Evrişimli Sinir Ağı (R-CNN) kullanılmıştır. Derin öğrenme tabanlı yaklaşım ile dişlerin bölgelerinin tespiti ve otomatik segmentasyonu için ilgi alanında bulunan dişlere ait piksel bilgisi sağlayan R-CNN ağı türünden olan Mask R-CNN ağı kullanılmıştır. Bu tez çalışmasının amacı, bitewing radyografileri üzerindeki dişlerin otomatik olarak segmentasyonunun gerçekleştirilerek diş bölgelerinin tespit edilmesidir. Çalışmada sinir ağının eğitimi için i9 10980XE işlemcili ve NVIDIA Quadro RTX 5000 ekran kartına sahip bir bilgisayar kullanılmıştır. Ek olarak çalışmada, derin öğrenme kütüphanelerinden olan Keras ve TensorFlow başta olmak üzere Python programlama diline ait kütüphaneler kullanılmıştır. Mask R-CNN sinir ağı ile kullanılan omurga ağı ise Artık ağlardan (Residual networks) olan ResNet-101 ağıdır. Diğer ağ modellerine kıyasla artık değerlerin sonraki modele eklenmesiyle oluşan ResNet omurga ağı, bu sayede klasik model olmaktan çıkmıştır. Deneysel çalışmalar sonucunda ise gerekli şekil, grafik ve çizelgeler ile daha detaylı analiz yapılmış ve elde edilen sonuçlar tartışılmıştır.
In this study, segmentation of teeth was provided on bitewing intraoral radiography images used in the treatment and diagnosis process of teeth that need to be examined in detail in the field of dental health. Bitewing imaging method is an ideal diagnostic method for detecting caries that cannot be seen directly in the mouth, especially caries at the interfaces of premolars and molars, and recurrent caries under restorations. Reporting of radiological images by experts is a manual, time-consuming and careful process. It is aimed to support these processes with a deep learning-based method so that they can be performed successfully and quickly. Legal permissions were obtained for the bitewing tooth images used in the study, and the dataset used in the study was obtained from the radiology image archive of the Ordu Dentistry Faculty, Department of Oral, Dental and Maxillofacial Radiology. In this study, Regional Based Convolutional Neural Network (R-CNN), which is a type of Convolutional Neural Networks (CNN), was used. The Mask R-CNN network, which is a type of R-CNN network that provides pixel information of the teeth in the region of interest, was used for the detection and automatic segmentation of the regions of the teeth with a deep learning-based approach. The aim of this thesis study is to determine the tooth regions by performing automatic segmentation of teeth on bitewing radiographs. In the study, a computer with i9 10980XE processor and NVIDIA Quadro RTX 5000 graphics card was used for the training of the neural network. In addition, libraries belonging to the Python programming language, especially Keras and TensorFlow, which are deep learning libraries, were used in the study. The backbone network used with Mask R-CNN neural network is ResNet-101 network, which is Residual networks. Compared to other network models, ResNet backbone network, which is formed by adding residual values to the next model, has thus ceased to be a classical model. As a result of the experimental studies, a more detailed analysis was made with the necessary figures, graphics and charts and the results were discussed."
In this study, segmentation of teeth was provided on bitewing intraoral radiography images used in the treatment and diagnosis process of teeth that need to be examined in detail in the field of dental health. Bitewing imaging method is an ideal diagnostic method for detecting caries that cannot be seen directly in the mouth, especially caries at the interfaces of premolars and molars, and recurrent caries under restorations. Reporting of radiological images by experts is a manual, time-consuming and careful process. It is aimed to support these processes with a deep learning-based method so that they can be performed successfully and quickly. Legal permissions were obtained for the bitewing tooth images used in the study, and the dataset used in the study was obtained from the radiology image archive of the Ordu Dentistry Faculty, Department of Oral, Dental and Maxillofacial Radiology. In this study, Regional Based Convolutional Neural Network (R-CNN), which is a type of Convolutional Neural Networks (CNN), was used. The Mask R-CNN network, which is a type of R-CNN network that provides pixel information of the teeth in the region of interest, was used for the detection and automatic segmentation of the regions of the teeth with a deep learning-based approach. The aim of this thesis study is to determine the tooth regions by performing automatic segmentation of teeth on bitewing radiographs. In the study, a computer with i9 10980XE processor and NVIDIA Quadro RTX 5000 graphics card was used for the training of the neural network. In addition, libraries belonging to the Python programming language, especially Keras and TensorFlow, which are deep learning libraries, were used in the study. The backbone network used with Mask R-CNN neural network is ResNet-101 network, which is Residual networks. Compared to other network models, ResNet backbone network, which is formed by adding residual values to the next model, has thus ceased to be a classical model. As a result of the experimental studies, a more detailed analysis was made with the necessary figures, graphics and charts and the results were discussed."
Açıklama
Anahtar Kelimeler
Bitewing radyografi, ağız içi görüntüleme, bilgisayar destekli teşhis, derin öğrenme, yapay sinir ağları, bölgesel tabanlı evrişimli sinir ağları, Mask R-CNN, örnek segmentasyon., Bitewing radiography, intraoral imaging, computer aided diagnosis, deep learning, artificial neural networks, region based convolutional neural networks, Mask R-CNN, instance segmentation.