Constrained Topological Mapping for Baseline Construction

This paper addresses the use of self–organizing maps for baseline construction in chromatograms. Unlike local techniques, the problem is seen in terms of global optimization: a straight and smooth path including sampled points with high significance for baseline membership is to be found. For their smoothing capabilities, and for reproducing the probability density function of the input, self–organizing maps allow for balancing between these demands accomplishing a kind of nonparametric weighted regression. The significances are determined from feature extraction and feature fusion at a local scale. Applying global optimization, robustness is achieved in two ways: First, the result will align to the position of the majority of significant points, and, second, a single false decision on the local level won‘t be able to change the course of the baseline completely ensuring comparability of peak measurements in similar chromatograms. Comparability, however, is essential for routine analysis which is the intended field of application.