Purchase this article with an account.
R.C. Bernardes, P. Baptista, J. Ferreira, L. Duarte, J. Cunha–Vaz; Earmarking Retinal Changes on a Color Fundus Photograph Time–Sequence . Invest. Ophthalmol. Vis. Sci. 2006;47(13):2644.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To automatically earmark changes on a time sequence of color fundus photographs and to allow the follow–on of individual retinal changes selected manually.
Color fundus photographs (50º field–of–view and 768 x 576 pixels) taken every 6 months during a period of 2 years (N = 5 images) by a Zeiss FF450 fundus camera were converted to gray–scale through Principal Component Analysis (PCA). This gray–scale image was then processed automatically so as to correct for non–uniform illumination, thus improving contrast and brightness. Fovea and optic disk locations as well as the retinal vascular tree were then determined in order to register the set of images, thus making it possible to compute the differences of each image to the baseline image in a total of N–1 difference–images. These can be considered as the components of a hyperspectral image, hence allowing PCA to be used to enhance the differences computed along the sequence. A color scheme was implemented to allow identification on a given instance of the sequence where the changes are coming from by giving a specific coloration to each visit. The computed changes can be mapped to any visit, showing their evolution over time. Any area can be individually followed regarding its size and/or shape.
A fully automated procedure was developed to allow mapping of the changes detected over a time–sequence of color fundus images onto a common reference image, thus allowing individual areas of detected changes to be monitored in terms of area and/or shape.
Changes automatically detected in color fundus images were successfully earmarked, despite different image conditions. The developed color scheme allows identification of the location, on the time–sequence, of the earmarked changes, as well as the areas where previous changes return to baseline status. Individual areas of detected changes can be followed in terms of their area and/or shape and these parameters are automatically plotted for graphical display.
This PDF is available to Subscribers Only