Classification of Retinal Detachment using Deep Learning through Retinal Fundus Images

In the field of Ophthalmology, Retinal Detachment is a serious condition that can lead to a severe loss of visual acuity if not treated on time. The early detection of Retinal Detachment can improve the chances of successful reattachment of the retina and the visual acuity, especially before macular involvement. The manual diagnostic of RD needed an experienced ophthalmologist and is quite time-consuming and labor-intensive. This work explores the role of deep learning for feature extraction, prediction, and classification of RD fundus images. In this paper, we have examined the performance of different neural networks like Googlenet, ResNet50, Densenet-201, InceptionV3, VGG16, and VGG19 for training and testing of the publically accessible database. Among these convolutional neural networks, ResNet50 yielded better statistical metrics in terms of accuracy, sensitivity, specificity, and AUC values of 99.16%, 99.59%, 98.26%, and 0.9991, respectively. This paper analyzes the diverse aspects for faultless prediction and classification of Retinal Detachment through color fundus images. This work helps to improve the ergonomic environment of clinicians in such a way it improves the treatment plan and supports the clinical decision.