ABSTRACT
The accuracy in classification while automatically detecting the retinal disease is always a critical design constraint in the field of ophthalmology. Deep nets found as the most promising technology than every other classification algorithms like SVM in terms of retinal disease detection accuracy. This paper proposes to compare the performance of two classifiers namely Probabilistic Neural Networks (PNN) and a deep learning technique called Convolutional Neural Networks (CNN) in differentiating the retinal fundus image as Normal, Diabetic Retinopathy (DR) and Age- related Macular Degeneration (AMD). The result indicates that deep nets are capable of providing better sensitivity, specificity with good accuracy. Hence, deep nets like CNN or MobileNets can be considered as an alternate to commonly used classifiers in the domain like PNN and SVM as well as its variants. This paper reveals the fact that deep learning techniques is the promising technology for image classification since CNN used for the counter comparison achieved an accuracy of 96.90% with sensitivity and specificity of 0.9863, 0.916 respectively.
