Glaucoma Detection using Transfer Learning

Abstract

Glaucoma is an eye disease that cause loss of eye if it is not treated, this disease is affecting a large number of people worldwide and creating lot of disturbances to the life of people. This disease will harm the optic nerve that sends visual data to eyes through brain, this optic nerve is the major reason for good vision. In this study, a suggested automated technique for detecting glaucoma from the fundus pictures depend upon pre-trained Convolutional Neural Network (CNN) types. The suggested approach not only aids in the early diagnosis of the glaucoma disease as well as promotes optometrists of taking quick decisions using accessible resources. The developed glaucoma detection approach made use of pre-trained ResNET50, VGG16, Xception, ResNET-101, InceptionV3, MobileNetV2, and EfficientNetB7 models. The Large-scale Attention based Glaucoma (LAG) dataset was used to assess the suggested technique. Using the LAG dataset, the ResNet50, VGG16, Xception, ResNET-101, InceptionV3, MobileNetB7, and EfficientNetB7 models produced satisfying results of 74%, 71%, 62%, 74%, 68%, 70%, and 74%, respectively. The ResNet50, ResNet-101, and EfficientNetB7 models were determined to be the best among these outcomes, achieving high accuracy with precision and recall of 74% and 74%, respectively.