Domain Adaptation for Bias Handling in the Detection of Diabetic Retinopathy

Diabetic Retinopathy (DR) is a major cause of blindness, triggered by long-standing diabetes. DR progresses slowly through these stages without proper screening and treatment and different lesions start appearing gradually in the eye, which distort the retina and harm the macula and, consequently, the vision. Regular screening and proper treatment after diagnosis are required to prevent this eye-threatening disease. Digital retinal imaging is usually used as a screening technique, and effective image processing techniques are needed to detect the different stages of DR and take prompt actions. The aim of this proposal is to network and establish collaborative research between different researchers in UK, India and the USA working on DR and related eye diseases, with the aim to develop an effective state-of-the-art automatic screening and classification system for DR and maculopathy using eye fundus images, in order to detect it at an early stage, so that to be used alongside human experts and save costs in hospitals. This is a complex multi-disciplinary task and, in order to accomplish it, collaborative work is needed between machine learning and computer vision researchers, software developers and clinicians/ophthalmologists. The long-term view is to extend the research collaboration to other eye-threatening diseases in addition to DR, such as glaucoma or Retinopathy of Prematurity, and develop effective Artificial Intelligence (AI) and computer vision-based automatic systems to detect these diseases in eye fundus images, with a performance comparable with that of a human expert/clinician.