Abstract
Purpose: :
To automatically compute differences over time from color fundus photography sequences in AMD patients’ eyes.
Methods: :
Color fundus photographs, 30º field-of-view centered on the fovea, were taken every 6 months during a 2-year follow-up period from 52 eyes of 52 patients. Images were converted to gray-scale through principal component analysis to capture the highest level of contrast from each color channel. Gray-scale images were thereafter processed to correct non-uniform illumination and normalize brightness. Vessel centerlines from the retinal vascular tree were segmented and used for image co-registration, based on a proprietary technique, to allow comparing any gray-scale image to the respective baseline and to compute differences. The process repeats for each image on the follow-up resulting in a sequence of difference-images. These differences can be added, after a specific coloration, and mapped, showing the progression of the detected changes over time accordingly to the defined color scheme.
Results: :
This fully-automated procedure allows to determine and to map changes detected over a time sequence of color fundus images. Specifically to these AMD patients’ eyes, this allowed to detect drusen size and confluence indicating turn over among other retinal changes from the difference-images.
Conclusions: :
The proposed method allows changes from color fundus photographs to be automatically determined, mapped and color coded accordingly to the baseline and was successfully applied to AMD patients’ eyes.
Keywords: image processing • age-related macular degeneration • retina