Using Self-Organizing Maps for object classification in Epo image analysis

Erythropoietin (Epo) is a hormone which can be misused for doping. The detection of its recombinant form (rEpo) involves analysis of Epo chemiluminescence images containing bands. Within a research project, granted by the World Anti-Doping Agency, we are developing the GASepo software to serve for Epo testing. For detection of the bands we have developed a segmentation procedure. Whereas all true bands are properly segmented, a relatively high number of artifacts is generated. The goal is therefore to separate the artifacts from the bands. In the paper an alternative classification method, based on self-organizing map, is proposed to solve the task of separation. The method performs well, when compared with other classification methods. In addition, it provides valuable insight into the properties of the data, their dependencies and their relevance for the classification task.