Diabetes is a disease that affects the structure of the eye and causes vision loss. It causes a wide variety of lesion types in the eye structure. It causes a wide variety of lesion types in the eye structure. These lesions on the retina images are symptoms of different diseases. The most well-known of these diseases is diabetic retinopathy. Detection of lesions is very important in early diagnosis and treatment of this ailment. In the study, a computer-assisted detection system based on Regional-Evolutionary Neural Networks has been proposed for the detection of lesions on the retinal images. With this proposed system, it is aimed to support the diagnosis and treatment of specialists working in the field of eye diseases. Retina images used in the study were obtained from STARE, DIARETDB0 and DIARETDB1 databases. 70% of the images in the databases used are devoted to education and 30% to test images. Regional-Evolutionary Neural Networks are designed in a region selector that allows multiple areas to be selected over a single image in order to tag educational images since they require a lot of data during the training phase. Retina images are trained on the cifar-10 pre-trained network, which is frequently used in deep learning practices. In the test operations performed at the end of the trainings, STARE, DIARETDB0 and DIARETDB1 databases successfully detected the lesion in the database with 91%, 98.53% and 93.01% accuracy, respectively.
: January 28, 2020
|APA||Uzun, S . (2020). Bölgesel-Evrişimsel Sinir Ağları ile Retina Görüntülerindeki Lezyonların Tespiti . Bilecik Şeyh Edebali Üniversitesi Fen Bilimleri Dergisi , 7 (1) , 34-46 . DOI: 10.35193/bseufbd.681195|