A Review of Preprocessing Methods for Retinal Fundus and OCT Images in Eye Disease Diagnosis

Abstract

Retinal imaging has become an essential tool in the field of ophthalmology, offering valuable insights into ocular health and supporting the diagnosis and monitoring of common vision-threatening diseases through non-invasive means (i.e. Fundus imaging and Optical Coherence Tomography (OCT)). While fundus imaging represents a two-dimensional view of the retinal surface given through visible light, the advantage of an OCT is the development of a cross-sectional view of the retina providing higher resolution details necessary to detect subtle changes within retinal layers. Preprocessing is also required to provide comprehensive image quality, reduce inter-image variability, and provide objective input to support the accuracy of automated diagnostic algorithms. In this report, I describe common and effective preprocessing techniques for both imaging modalities: 1. fundus image preprocessing includes improvement to image quality, vessel and structure enhancement, color normalization, noise reduction, and augmentation of fundus images; 2. OCT image preprocessing includes denoising, flattening, motion correction and cropping, and intensity normalization of OCT images. With this study, the various preprocessing types of modalities can improve clinical utility and efficiency, and delineate a process where machine learning (ML) and computer vision can be incorporated into the diagnostic process.