Computing Degree Show 2018

Synthesis of realistic retinal fundus images by deep learning

Brian Stevenson: Synthesis of realistic retinal fundus images by deep learning is generating synthetic images (phantom) that can be used as part of a dataset for research. The aim of my project was to improve the vascular structure by improving the code that generates the blood vessel network in the retina via matlab. To allow the vessels to look more realistic and have a realistic path of what the vascular structure should look like as it currently is not represented within the phantom. The method used is a deep learning approach with real and synthetically generated examples. This will allow Beijing to use the software created by myself as part of vampire in collaboration with Singapore where this will create a library of real and synthetically generated images to be used to diagnose problems within the retina more quickly by comparison. The application has a GUI that allows a background image to be displayed first of all and then generating the vascular. The current problem is synthetically generated vascular in the past has not been very realistic, so we hope to have achieved more realistic vascular.