Deep Learning Approach to Enhance Accuracy for Early Detection of Glaucoma

Abstract

Diabetes is a medical disorder when the blood sugar (glucose) level cannot be controlled by the body. This can occur if the body can't properly use the insulin it produces or if the body doesn't produce enough insulin. Diabetes can lead to major health issues and increase your chance of developing a number of eye illnesses if it is not properly managed. The advancement of machine learning algorithms has made early detection of various eye illnesses using an automated method significantly more advantageous than manual detection. The ocular illness that lead to visual loss is Glaucoma which do not have any symptoms. Early detection can help to reduce disease-related vision loss. This study proposes a segmentation using UNet model (which is a U-shaped encoder-decoder network architecture, which consist of four encoder blocks and four decoder blocks that are connected via a bridge) on fundus images followed with data augmentation. The CNN (Convolution Neural Network) model is then trained using pre-processed fundus image. The proposed model was using IEEE dataset named REFUGE (Retinal Fundus Glaucoma Challenge). In an evaluation after 100 epochs, the accuracy is 98%. The proposed model outperforms existing deep learning model for early detection of glaucoma.