Purchase this article with an account.
A. R. Chaudhry, C. Bellmann, E. Parra-Denis, J.-C. Klein; Classification of Fundus Autofluorescence (FAF) Patterns in Age-Related Macular Disease (AMD). Invest. Ophthalmol. Vis. Sci. 2008;49(13):1856.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop an automatic analysis method for the classification of FAF patterns that may be important in computer assisted diagnosis of AMD as the major cause of legal blindness in people over 55 years of age.
The software for automatic classification of FAF patterns was developed from an annotated database including 200 FAF images (acquired with HRA2) of patients suffering from different stages of AMD.
A first version of the software is now available. It includes two different processing steps:1. Region of Interest (ROI) extraction: The objective of ROI extraction is to retain image zones with significant FAF signal, which is caused by lipofuscin accumulation in retinal pigment epithelial cells. Blood vessels and papilla, which are deficient in FAF signal are detected and not taken into account in pattern evaluation process.2. Feature extraction and classification: Several FAF patterns may appear simultaneously on the same retinal image. Therefore, the textural patterns are calculated locally within concentric circular zones centred on macula by morphological and linear operators.The first classification phase of FAF patterns is accomplished by clinical supervised cataloging into several classes. Annotated images served for the tuning of classifier and selection of features, which are crucial for an efficient classification in FAF imaging.In a second phase, an unsupervised classification is achieved using the more efficient previously selected features. The system forms clusters or "natural groupings" of the input FAF patterns.The initial evaluation performed on the database show a close relationship between subjective clinical visual annotation and automatic classification.Conclusion This automatic classification of FAF images will provide a detailed and more discriminant classification for the diagnosis, follow-up patients and evaluation of novel treatments in AMD.
This PDF is available to Subscribers Only