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

Automatic detection of retinal disorders is gaining considerable attention with the emergence of deep learning. Ophthalmologists primarily use color fundus photographs to examine the human retina and diagnose the abnormalities. As there is a surge in the number of visual impairments, an AI-enabled retina screening system can expedite the retina examination process. Existing works in this direction are primarily focused on either segmentation or classification. Furthermore, the majority of the works are implemented using preprocessed good quality fundus images. In reality, however, the quality of color fundus images is degraded due to the illumination inhomogeneity and low contrast issues. Thus, there is a need to develop an end-to-end fundus image analysing system. Steering in this direction, the proposed work attempts to analyze the performance of semi-supervised Generative Adversarial Networks (GANs) for the classification of retinal fundus images into multiple categories. Besides, the nonlocal retinex framework is applied to enhance the quality of fundus images without over-smoothing the edges. The large data set of raw fundus acquired from multiple Eye hospitals and released in public domain is used to implement the proposed work. The results obtained are compared with the transfer learning method, and an average accuracy of 87% is obtained. It suggests that the semi-supervised GANs can be potentially used to classify heterogeneous retinal disorders. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd 2021.