Understanding the biomarkers of retinal disease using deep learning

Abstract

The retina has biomarkers not only for ophthalmic disease but also for diseases and conditions across the entire body. In this paper, I focus on retinopathy of prematurity (ROP), a proliferative vascular disease that can cause blindness in prematurely born infants. To prevent further disease progression and vision loss in infants with ROP, early and accurate detection of plus disease is crucial. In current practice, clinicians compare retinal fundus photographs to a reference standard image in order to detect plus disease for severe ROP. This process can be highly qualitative, subjective, and variable. Furthermore, some clinical environments may lack clinicians with the expertise to diagnose these diseases. I am to address these shortcomings in current clinical diagnosis of ROP by using deep learning methods to automatically extract biomarkers of disease without human intervention. Since ROP is primarily present in prematurely born infants, I also attempt to predict and analyze gestational and postmenstrual age, and how disease predictions vary from healthy to affected infants.