Classification of proliferative diabetic retinopathy using deep learning

摘要

Diabetic retinopathy (DR) is a disease with different degrees of severity. The patterns that denote the presence of the disease are varied; when the disease is in low level severity it's hard to find the features that represent the presence of DR. In this paper, two convolutional neuronal networks (CNN) are proposed to classify proliferative diabetic retinopathy. For the CNNs training and evaluation the Kaggle public data base that has around 35,000 training images was used. The obtained classification accuracy using the two CNNs architectures was of 96.37% and 97.38% respectively.