Fundus Image-Based Macular Edema Detection Using Convolutional Neural Network

Abstract

AbstractMacular disorders are a set of diseases that damage the macula in retina. They give rise to distorted vision and in some cases may even lead to visual impairment. Macular edema (ME) is one of the most crucial types among macular disorders and it is caused by an accumulation of fluid under the macula. Various techniques have been developed till date to detect macular edema; however, the detection of macular edema alone is not enough. It is also essential to give the correct diagnosis and appropriate medical treatments based on the severity of edema. In this paper, an automated system for the detection of macular edema from fundus images has been discussed. The system uses features of convolutional neural networks (CNN), a sophisticated deep learning module for classification of exudates. The information extracted from the following input of a fundus image is distinguished with the training data provided on the CNN module to grade the macular edema. The edema, at an early stage, is categorized as a “mild case of ME,” a slightly advanced stage as a “moderate case of ME” and at a highly advanced stage as a “severe case of ME”; based on these categorizations, appropriate medications and treatments are suggested to the patients.KeywordsConvolutional neural network (CNN)Macular edemaOptimizerLearning rateEpoch