Automated laterality determination in Ultra-Widefield Scanning Laser Ophthalmoscopy

Peter Wakeford, Enrico Pellegrini, Jano van Hemert, Ik Siong Heng

Abstract: A deep learning method to determine the laterality of ultra-widefield retinal images is presented. On a test set of 411 images, a 98.3% accuracy was achieved. Saliency maps show that the network activates on the same features which a human reviewer would use to determine laterality.

Keywords: Image processing, deep learning, retinal imaging

Author Affiliation: University of Glasgow, Optos plc (A Nikon company)