Deep Learning Based Models for Detection of Diabetic Retinopathy

Abstract

Diabetic retinopathy (DR) is an important disease that occurs because of damage to the retinal blood vessels in the human eye due to diabetes and causes blindness. If diagnosed correctly, the treatments to be applied increase the possibility of preventing vision loss or blindness. This study aims to present an evaluation of deep learning methods to detect diabetic retinopathy from retinal images. In this direction, the VGG16 model was considered, and two different versions of this model were obtained by making improvements. Besides, a model has been proposed, the first layers are dense, the next layers have decreasing convolution, and have fewer layers. According to the results, the VGG16 model, which reached 75.48% accuracy, reached 76.57% accuracy due to the dropout layer added to the classification layers, and 77.11% accuracy due to the dropout layer added to all blocks. The highest accuracy was obtained in the proposed model with 81.74%.

Keywords

• artificial intelligence
• deep learning
• detection
• diabetic retinopathy