The Conception of Fundus Multi-Disease Dataset (FMDD) using Multi-Spectral Generative Adversarial Networks

The World Health Organization (WHO) reports that 2.2 billion people are affected by visual impairment. Early detection and diagnosis of ocular pathologies can help predict visual impairment. Over the past twenty years, many fundus image datasets have become publicly available due to technological advances. These datasets have primarily focused on the detection of common ocular pathologies such as age-related macular degeneration, diabetic retinopathy, and glaucoma, and recent research in fundus diseases has highlighted the importance of detecting multiple fundus diseases. Existing datasets such as ARIA and RFMiD mainly contain images of the most common pathologies and very few images related to rare pathologies. The existing public datasets have problems in multi-disease classification, such as less data under some under-represented diseases, low-quality photos, and class imbalance among several classes. The main objective of our research is to construct a Fundus Multi-Disease Dataset (FMDD) with 20 courses of ocular diseases from publicly available datasets and with the application of Multi-Spectral Generative Adversarial Networks (MSGANs). The resulting dataset is balanced for all image classes.