Peak recognition imitating human judgement

Peak integration is still a major source of error in analytical techniques such as chromatography (LC and GC), aapillary electrophoresis (CE), spectrosocpy, and electrochemistry. If the baseline is complex, e.g. because of matrix effects, or if the peak shape is irregular, e.g. because of peak tailing, the results are often not satisfactory when classical procedures are used. These shortcomings arise because of the stepwise appearance of the chromatogram. An algorithm that copies the human method of considering baseline and peaks as a whole has already been introduced. Here the use of a straight line as a baseline model led to an improvement in several instances. The baseline is, however, usually not exactly straight and rigid. A baseline model with flexible properties is more advantageous. Thus the smoothing cubic spline function is applied in this work. Here the rigidity can be controlled by use of a parameterp k. The prediction interval of the spline is used for iterative distinction between baseline and peak regions. Afterwards straightforward optimization of the peak boundaries is applied. More than 50 series of consecutive injections of the same sample (n=40 on average) were used to test the performance of this procedure. The same raw data have been integrated by means of the algorithm described here and by use of commercially available software. The reproducibility of the main component peak are within the series was taken as a measure of integration quality. Typically the new procedure reducesRSD % by approximately 33% (e.g. from 1.5% to 1.0%). The improvement is even more impressive for difficult samples with complex matrices, e.g. blood plasma or polymer excipients. for such samples improvements of up to a factor of 6 are obtained.