Classification of Optical Coherence Tomography Images for the Diagnosis of Retinal Diseases using Deep Learning Methods

Esra Urmamen; Sabri Koçer

Article Details

Classification of Optical Coherence Tomography Images for the Diagnosis of Retinal Diseases using Deep Learning Methods

Authors:

Esra Urmamen | Sabri Koçer

Publisher:

Eskişehir Osmangazi University

Main Topics:

Bilgisayar Yazılımı

Journal:

Eskişehir Türk Dünyası Uygulama ve Araştırma Merkezi Bilişim Dergisi Vol.4 Issue 3, pp.22-28

Document Type:

Article

Publication Date:

12/2023

Document Language:

Turkish

Document ID (Org No.):

2001296204

pISSN:

2687-606X

Keywords:

Abstract:

 The retina is light-sensitive neural layer that enables vision and perception of colors. Distortions in the retina can negatively impact peoples quality of life. Such distortions can lead to serious problems, including blindness and permanent damage to the retina. With the advancement of technology in the treatment of retinal diseases, the use of computer-aided diagnosis systems has become increasing common. Early diagnosis and treament can prevent permanent damage to the retina and help patients retain their vision. As technology has progressed, cameras and computer-aided diagnosis systems have been widely adopted. Retinal images obtained using OCT devices enable experts to make more accurate and early diagnoses. In this study, transfer learning methods, specifically InceptionV3,Xception, and the proposed Convolutional Neural NetworkCNN model, were compared for classifying retinal diseases. The Xception network achieved an accuracy of 95.36, while the Inception network achieved an accuracy of 98.2. The proposed CNN architecture achieved an accuracy of 97.51. The proposed architecture has obtained more successful results in the classification of diabetes and normal diseases based on diseases architecture compared to other methods.

 The retina is light-sensitive neural layer that enables vision and perception of colors. Distortions in the retina can negatively impact peoples quality of life. Such distortions can lead to serious problems, including blindness and permanent damage to the retina. With the advancement of technology in the treatment of retinal diseases, the use of computer-aided diagnosis systems has become increasing common. Early diagnosis and treament can prevent permanent damage to the retina and help patients retain their vision. As technology has progressed, cameras and computer-aided diagnosis systems have been widely adopted. Retinal images obtained using OCT devices enable experts to make more accurate and early diagnoses. In this study, transfer learning methods, specifically InceptionV3,Xception, and the proposed Convolutional Neural NetworkCNN model, were compared for classifying retinal diseases. The Xception network achieved an accuracy of 95.36, while the Inception network achieved an accuracy of 98.2. The proposed CNN architecture achieved an accuracy of 97.51. The proposed architecture has obtained more successful results in the classification of diabetes and normal diseases based on diseases architecture compared to other methods.

 The retina is light-sensitive neural layer that enables vision and perception of colors. Distortions in the retina can negatively impact peoples quality of life. Such distortions can lead to serious problems, including blindness and permanent damage to the retina. With the advancement of technology in the treatment of retinal diseases, the use of computer-aided diagnosis systems has become increasing common. Early diagnosis and treament can prevent permanent damage to the retina and help patients retain their vision. As technology has progressed, cameras and computer-aided diagnosis systems have been widely adopted. Retinal images obtained using OCT devices enable experts to make more accurate and early diagnoses. In this study, transfer learning methods, specifically InceptionV3,Xception, and the proposed Convolutional Neural NetworkCNN model, were compared for classifying retinal diseases. The Xception network achieved an accuracy of 95.36, while the Inception network achieved an accuracy of 98.2. The proposed CNN architecture achieved an accuracy of 97.51. The proposed architecture has obtained more successful results in the classification of diabetes and normal diseases based on diseases architecture compared to other methods.

 Retina, görmeyi sağlayan ışığa ve renklere duyarlı ağ tabakasıdır. Retinadaki bozulmalar insanların yaşam kalitesini düşürmektedir. Retinada meydana gelen hasarlar körlüğe varan ciddi sorunlara sebep olabilmekt e ve retinada kalıcı hasarlar meydana gelebilmektedir. Retinal hastalıkların tedavisinde gelişen teknoloji ile birlikte bilgisayarlı tanı sistemlerinin kullanımı oldukça yaygınlaşmıştır. Erken teşhis ve tedavi edilmesi retina da oluşabilecek kalıcı hasarla rı ve hastaların görme yetisini kaybetmesini önlemektedir Teknolojinin ilerlemesiyle birlikte fotoğraf makineleri ve bilgisayarlı tanı sistemleri oldukça yaygın kullanılmaya başlanmıştır. OCT cihazları kullanılarak elde edilen retinal görüntüler uzmanların daha doğru ve erken teşhis koymalarını sağlamaktadır. Bu çalışmada, retinal hastalıkların sınıflandırılması için transfer öğrenme yöntemlerinden InceptionV3, Xception ve önerilen Evrişimsel Sinir Ağı ESA modeli karşılaştırılmıştır. Xception ağında 95.36 oranında doğruluk değerine, Inception ağında ise 98.2 oranında doğruluk oranı elde edilmiştir. Önerin ESA mimarisinde  97.51 oranında doğruluk oranı elde edilmiştir. Önerilen mimari hastalık bazında diyabet ve normal hastalıkların sınıflandırılmasında diğer yöntemlerden daha başarılı sonuçlar elde etmiştir.

Article Details

Classification of Optical Coherence Tomography Images for the Diagnosis of Retinal Diseases using Deep Learning Methods

Authors:

Esra Urmamen | Sabri Koçer

Publisher:

Eskişehir Osmangazi University

Main Topics:

Bilgisayar Yazılımı

Journal:

Eskişehir Türk Dünyası Uygulama ve Araştırma Merkezi Bilişim Dergisi Vol.4 Issue 3, pp.22-28

Document Type:

Article

Publication Date:

12/2023

Document Language:

Turkish

Document ID (Org No.):

2001296204

pISSN:

2687-606X

Optical Coherence Tomography

Deep Learning

Convolutional Neural Network

Transfer Learning

Retinal OCTOptik Koherans Tomografi

Derin Öğrenme

Evrişimli Sinir Ağı

Transfer Öğrenme

Issues

Article Details

Authors: Esra Urmamen | Sabri Koçer

Publisher: Eskişehir Osmangazi University

Main Topics: Bilgisayar Yazılımı

Journal: Eskişehir Türk Dünyası Uygulama ve Araştırma Merkezi Bilişim Dergisi Vol.4 Issue 3, pp.22-28

Document Type: Article

Publication Date: 12/2023

Document Language: Turkish

Document ID (Org No.): 2001296204

pISSN: 2687-606X

Issues

Authors:

Esra Urmamen | Sabri Koçer

Publisher:

Eskişehir Osmangazi University

Main Topics:

Bilgisayar Yazılımı

Journal:

Eskişehir Türk Dünyası Uygulama ve Araştırma Merkezi Bilişim Dergisi Vol.4 Issue 3, pp.22-28

Document Type:

Article

Publication Date:

12/2023

Document Language:

Turkish

Document ID (Org No.):

2001296204

pISSN:

2687-606X