gms | German Medical Science

102. Jahrestagung der DOG

Deutsche Ophthalmologische Gesellschaft e. V.

23. bis 26.09.2004, Berlin

Computer-assisted automated analysis of digital cSLO fundus autofluorescence images in geographic atrophy due to AMD

Meeting Abstract

  • corresponding author S. Schmitz-Valckenberg - Dept of Ophthalmology, University of Bonn
  • A. Deckert - Institute for Medical Biometry and Applied Informatics (IMBI)
  • J.J. Jorzik - Dept of Ophthalmology, University of Heidelberg
  • A. Bindewald - Dept of Ophthalmology, University of Bonn
  • F.G. Holz - Dept of Ophthalmology, University of Bonn
  • U. Mansmann - Institute for Medical Biometry and Applied Informatics (IMBI)
  • FAM-Study Group

Evidenzbasierte Medizin - Anspruch und Wirklichkeit. 102. Jahrestagung der Deutschen Ophthalmologischen Gesellschaft. Berlin, 23.-26.09.2004. Düsseldorf, Köln: German Medical Science; 2004. Doc04dogP 161

Die elektronische Version dieses Artikels ist vollständig und ist verfügbar unter:

Veröffentlicht: 22. September 2004

© 2004 Schmitz-Valckenberg et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen ( Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.




Fundus autofluorescence (FAF) imaging using confocal scanning laser ophthalmoscopy (cSLO) has been shown to be superior to fundus photography for delineating areas of geographic atrophy (GA) of the retinal pigment epithelium. Computer-aided tools can be helpful for monitoring growth of GA over time. Based on our previous work we have assembled and evaluated an advanced customized software for automated detection and quantification of GA areas.


Using a region-growing algorithm and an algorithm for the elimination of interfering vascular structures, atrophic areas were measured by method C on FAF images (HRA: exc: 488nm, em. > 500nm) of 34 GA patients by two independent readers who had been analysed the same images before (Method B, Graefe's 2002). Agreement between the observers and between both methods was assessed.


The new developed algorithm reliably identifies vascular structures that interfere with the GA. Tools such as contrast enhancing, sensitive threshold adjustment and relocatable convex and radials hulls facilitate fine-tuning of the actual segmentation of GA areas. The agreement between the readers was -0.108±0.489 for method B and 0,121±0,278 for method C using the Bland-Altman-test. The Friedman-test showed a global interaction between methods and readers (p<0.001).


Computer-assisted methods are necessary to reduce procedure-dependent variabilities while analysing atrophic areas of GA patients in longitudinal studies and interventional trials for slowing down progression of atrophy. Here, we could show several important aspects for the improvement of measuring atrophic areas and could successfully tested adjusted algorithms. The further development of the FAF images analysis of GA patients demands a combination of these algorithms and an integration of different concepts.