A Study of Bilateral Symmetry in Color Fundus Photographs

Abstract

We have previously shown that there is a high degree of bilateral similarity in the central retinal blood vessels (CRBVs), which are responsible for supplying blood to retinas and can be used as a strong biometric. We have also shown that a side-independent retina verification system can be developed based on the bilateral similarity in CRBVs. In this paper, we perform a similar investigation for color fundus photographs since color fundus photographs are much richer representations of retinas than CRBVs. We investigate whether the color fundus photographs of the left and right retinas possess strong enough bilateral symmetry so that we reliably tell whether a pair of the left and right retinas belong to a single subject. We evaluate and analyse the performance of both human-and deep neural network-based bilateral verification by experimenting on color fundus photographs of two publicly available data sets.

Diagnosis of Multiple Sclerosis by Detecting Asymmetry Within the Retina Using a Similarity-Based Neural Network

“…In Section III-A, we discuss how weight sharing may effect the performance of the system, depending on whether it relies on symmetry or similarity for its analysis. Another similar study uses an SNN trained on fondus images, along with a test group, to determine similarity in pairs of eyes [23], with the SNN substantially outperforming human volunteers. This highlights the potential value of incorporating such models into clinical settings.…”