Deriving Baseline Detection Algorithms from Verbal Descriptions

The presented strategy of automatic baseline detection in chromatograms combines fuzzy logic and neural network approaches. It is based on a verbal description of a baseline referring to a 2D image of a chromatogram instead of a data vector. Baselines are expected to touch data points on the lower border of the chromatogram forming a mainly horizontal and straight line. That description has been translated into a couple of algorithms forming a two–stage approach first proceeding on a local, and second, on a global level. The first stage assigns a value regarded as the degree of baseline membership or significance to each data point; the second uses a global optimization strategy for coordinating these significances and for producing the final curve, simultaneously. The statistical stability of the proposed approach is superior to known approaches, while keeping the computational effort low. 1 Motivation Baseline definition. Chromatographic separation is a widely used technique for quantifying mixtures of substances. The mixture is decomposed into a sequence of their individual components resulting in an intensity vs. time profile – the chromatogram – as shown in figure 1. Substances can be distinguished by the positions of their corresponding peaks; peak area is a measure of the amount of a substance [2]. An ideal chromatogram consists of well–separated Gaussian peaks on a flat baseline. But usually, peak overlap, negative peaks and drifts of varying sign can occur, and even disturbances of the kind shown in figure 3 can be recorded. Thus, for correct peak measurement, and therefore, for correct quantification, a correcting signal, the baseline, is searched for ignoring peak overlap, separating positive peaks from negative ones, and following drifts and ruptures. Other approaches. Most automatic baseline detection strategies (e.g. [3]) assume a mainly horizontal and straight curve. Following this, baseline candidates are those points having small or zero slope. Thus, minima between overlapping peaks and the tips of negative peaks will be marked by mistake. This problem is usually handled by introducing a threshold upon the drift: the line connecting two adjacent points marked as baseline members may not exceed a given slope. Though including additional data points inthe decision, the strategy remains local. Threshold criteria on slope and drift are critical in multiple ways: first, a single local decision on baseline membership can change the further course of the baseline completely such that peak measurements of two similar chromatograms will possibly not be comparable; second, in general, a constant threshold will not be applicable to the whole chromatogram: with the threshold being too low, the baseline cannot follow the chromatogram, otherwise, it runs into peak groups. Besides, even data points with higher slope can be baseline members [4]. 2 Concept