Budai, Attila; Michelson, Georg; Hornegger, Joachim: Vessel Segmentation on Retinal Fundus Images at The Association of Research in Vision and Ophthalmology (Konferenz) in Fort Lauderdale, FL, USA (03.5.)

Budai, Attila; András Lassó; Kálmán Palágyi; Tamás Ungi; Gábor Novák: Blood Vessel Segmentation on Coronary Angiograms In: Feußner, Hubertus (Eds.) Proceedings of the 5rd Russian-Bavarian Conference on Biomedical Engineering (5rd Russian-Bavarian Conference on Biomedical Engineering München 1-4.7.2009) 2009, pp. 21-24

Budai, Attila; Michelson, Georg; Hornegger, Joachim: Multiscale Blood Vessel Segmentation in Retinal Fundus Images In: Meinzer, Hans-Peter; Deserno, Thomas Martin; Handels, Heinz; Tolxdorff, Thomas (Eds.) Bildverarbeitung für die Medizin 2010 - Algorithmen, Systeme, Anwendungen, (Bildverarbeitung für die Medizin 2010 - Algorithmen, Systeme, Anwendungen, Aachen 14-16.3.2010) Heidelberg : Springer 2010, pp. 261-265 - ISBN 978-3-642-11967-5

Current research / employment

The eye is a unique organ where the human vasculature is visible in-vivo and non-invasive. The most common modality for eye examinations is the so called fundus camera taking photographs of the eye background through the pupil. These images are used to diagnose eye diseases as well as to examine the human vascular system. The evaluation of these images is made usually by experts manually by looking at each of the images one by one and searching for abnormalities and diseases. In case of multiple patients it can take several hours to check for irregular structures.

The aim of my research project is to develop an algorithm for automatic analysis of fundus photographs with minimal human interaction to support the ophthalmologists. As a first step the vessel tree is segmented in the image. The segmentation algorithm is using a multi-resolution approach to speed-up a state-of-the-art vessel enhancement algorithm, and uses a hysteresis thresholding to increase its sensitivity.

The segmented images are part of the input for further methods to measure vessel properties like thickness and tortuosity (the waviness of the vessels). Some methods are analyzing the eye background, and the blood vessels would corrupt the measurements. In this situations the segmentation results can provide a mask of irrelevant structures. An example can be a method to calculate the macular pigment density changes.

Figure: (left) Common fundus image: Red channel usually oversaturated; (right) Visualized tortuosity of vessels.