A Semi-supervised Generative Adversarial Network for Retinal Analysis from Fundus Images

Classification of Multiple Retinal Disorders from Enhanced Fundus Images Using Semi-supervised GAN

“…Here N, the number of categories is set to 5. The depth of the discriminator network in the proposed work is increased slightly compared to our previous work on semi-supervised GAN [7]. This is done empirically on the ODIR data set.…”

“…One is to classify fundus images into 5 different categories ( C S ), while the other branch is to distinguish between real and fake images ( D S ). A notable contribution of the proposed work is to extend the number of classes from 2, (normal/abnormal as in [7]) to 5, where the classes are Glaucoma, DR, Normal, AMD, and others. The loss function of classifier branch C S is as given below:…”

“…The classification metrics obtained for the proposed model is presented in Table 2. Our preliminary work on semi-supervised GAN [7] was able to achieve 54% accuracy when only a subset of fundus images from DR and Glaucoma was considered. However, in this work, the finetuned model is trained on the augmented and enhanced larger sets of fundus images.…”

“…the objectives of proposed approach is to enhance the quality of fundus images using perceptually inspired nonlocal retinex frameworks. The proposed approach is an extension of our previous work on fundus image analysis using semi-supervised Generative Adversarial Network (GAN)s [7]. It was noted that the deep learning model discussed in [7] under-performed for a specific type of public data set.…”

“…The proposed approach is an extension of our previous work on fundus image analysis using semi-supervised Generative Adversarial Network (GAN)s [7]. It was noted that the deep learning model discussed in [7] under-performed for a specific type of public data set. Therefore, to boost the model's performance, a preprocessing module was appended to the semi-supervised model.…”