A Deep Learning Approach for Ocular Disease Detection

Abstract

The early identification of ocular disease (OD) detection is essential in preventing complete blindness. Although much information is available online, ophthalmologists have valuable information to diagnose the condition. Still, it also creates many challenges due to the increase in the variation in fundus images. The diagnosis of disease using hand-crafted techniques on a manual basis is time-consuming. It is unsuitable in countries like India, where the blind population is approximately 16 million. In this paper, we proposed an approach to diagnose OD automatically, where detection was done in two steps. The MobileNet architecture is used for feature extraction since it is suitable for smartphone/iPhone users those does not have computer systems at home. The architecture is faster than other available architectures, such as VGG and RESNET. The network was trained on the data of over 3500 patients and tested over 1500 patients giving an accuracy of 95.68% when validated.