Spherical SOM and its Application to Fundus Image Analysis

Abstract

A new cluster analysis using spherical SOM has been proposed to create a more accurate representation of the data by removing the "border effect". The cluster method is able to visualize a multi-dimensional dataset as a graphical object and to extract a tree structure from the object as a dendrogram. In this paper, to better understand the potential effectiveness of the method, the cluster method using spherical SOM has been applied to real medical treatment data and has been compared with the results of the other three approaches for clustering, which are Learning Vector Quantization, Support Vector Machine and Discriminant Analysis. The results of cluster analysis for real medical treatment data confirmed that the cluster analysis using spherical SOM was accuracy more than the equal to the other approaches.