Evolutionary Generative Adversarial Networks applied to computer-assisted diabetic retinopathy diagnosis

• Grant number: 19/00585-5
• Support Opportunities: Regular Research Grants
• Duration: August 01, 2019 - July 31, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Convênio/Acordo: Australian Technology Network of Universities (ATN)
• Mobility Program: SPRINT - Projetos de pesquisa - Mobilidade
• Principal Investigator: João Paulo Papa
• Grantee: João Paulo Papa
• Principal researcher abroad: Dinesh Kant Kumar
• Institution abroad: RMIT University, Melbourne, Australia
• Host Institution: Faculdade de Ciências (FC). Universidade Estadual Paulista (UNESP). Campus de Bauru. Bauru , SP, Brazil
• Associated research grant: 14/12236-1 - AnImaLS: Annotation of Images in Large Scale: what can machines and specialists learn from interaction?, AP.TEM

Abstract

Undiagnosed diabetic retinopathy (DR) leads to vision impairment and blindness, and its early detection can significantly reduce severe vision impairment by 50%. The presence of exudates in fundus images is one of the first signs of such a disease, but their manual detection is examiner-dependent and time-consuming. In this proposal, we aim at coping with the lack of data concerning computer-assisted DR detection by synthesizing retina images through Generative Adversarial Networks (GANs). The proposal will introduce the concept of Evolving GANs, where evolutionary optimization will be employed to either fine-tune GAN hyperparameters and to learn composite loss functions. The Brazilian team will be in charge of the machine learning side; meanwhile the Australian team will take care of the dataset and the knowledge about the problem. (AU)