Gennotate

Abstract

Common eye diseases like cataract, glaucoma, and diabetic retinopathy often lead to blindness or other complications. However, the automated disease diagnostics becomes a challenging task due to the data scarcity caused by the patients’ privacy concerns. Also, the process of manually annotating the data is a very tedious task for the doctors. For addressing these issues, gennotate focuses on generating high quality Optical Coherence Tomography (OCT) images through Generative Adversarial Networks (GANs). The generated images can be used by AI professionals to train and test the diagnostic algorithms, thus addressing the issue of data scarcity. The quality of the images generated has been verified by an ophthalmologist who has ensured the high quality of images. Moreover, a classification model is trained to evaluate the difference between the accuracies of real and generated images. The issue of doing tedious manual annotations is being resolved by automating the semantic segmentation of Optical Coherence Tomography (OCT) images through SegFormer model. For evaluating the quality of the images segmented, the difference in the Mean Intersection over Union (IOU) of real and generated images has been compared. In addition to that, gennotate provides a web application that enables real-time Optical Coherence Tomography (OCT) image generation, segmentation, classification and an editor for manual annotation of images, thus innovating the field of automated disease diagnostics.