The quantization of the retinal vein contour for the automatic diagnosis of arteriosclerosis by using SOM

Abstract

One of the major indicators in the severity diagnosis of arteriosclerosis is the shape pattern of the vein around crossing the artery in the retina. In this study, for automation of the diagnosis, we calculate feature vectors by quantization of the retinal vein contour around the crossing. The quantization requires to normalize the vein contour and to correspond the feature vectors order. We propose to apply the Self-Organizing Maps(SOM) to quantize compatibly with the normalization and the feature vectors order for preprocessing of the automation diagnosis. In the experiments, the proposed method was applied to the vein contour, that extracted from fundus images. The evaluation of the experimental results from the viewpoints of the severity diagnosis and the quantization errors shows effectiveness of the proposed method.