Multimodal Image Registration of Retinal Images

Frédéric Zana


Introduction

Registration is an essential task in medical imaging, but this is particularly true in retinal analysis related to diabetes. The evolution of the quality of vision is a direct consequence of the deterioration of the vascular tree caused by diabetes. Most diagnosis require comparison of various images provided by different sources, spaced in time. Furthermore vascular abnormalities can deeply modify the aspect of the eye fundus. We are faced with a problem of multimodal and temporal registration with possible occlusion effects. In order to solve such a difficulty, we have chosen to detect the vascular tree and we use a registration algorithm that deals with occlusion effects.


The algorithm


Results

left: angiographic image right: negative of the corresponding green image. Red dots (above) are used for registration. The result appears below, the detection of the vessels appears in white for the angiographic image, in black for the green image and in grey where it is superimposed.

.

left: before aggravation (vessels in white below) right: two years later (vessels in black below). below: registration

.

left: before aggravation (vessels in white below) right: two years later (vessels in black below). below: registration

.

left: angiographic image right: corresponding green image below: registration

.

left: angiographic image with occlusion du to over-exposure right: green image two years later below: registration


    Frédéric Zana
    born the 9th of May 1972
    35 rue Saint-Honoré
    77 305 Fontainebleau, FRANCE