High-resolution fragmentary decomposition-a model-based method of non-stationary electrophysiological signal analysis.

摘要

Fragmentary decomposition (FD) is a recently developed method of non-stationary electrophysiological signal analysis addressed to mass potentials, such as electromyogram (EMG), event-related potential (ERP), evoked potential, electroencephalogram (EEG), electroretinogram, etc. Being supported by the generally accepted physiological notion that a peak is a functionally meaningful component of a mass potential, FD provides a way to avoid averaging and, instead, quantifies the component composition of complex electrophysiological signals directly from single-trials. The major computational procedures of FD include adaptive segmentation, the frequency domain component identification, and creation of the signal model as a linear aggregation of multiple components, with the generic mass potential (GMP) being the universal component template. This paper presents an improved, high-resolution FD technique which allows the resolution of overlapping sub-components and supports each identified component by an individual model. On the basis of this methodological innovation, we define two fundamental categories of multi-peak component waveforms: complex components (CC), comprised of multiple sub-components (GMPs), versus monolithic components (MC), involving a single GMP. We show that quantification of MCs and CCs from single-trial eyeblink EMG and single-trial ERP provides a more comprehensive analysis of these signals. Given single-trial eyeblink EMG, we find that the stimulus elicits strong though short-term (phasic) effects on MCs and moderate but long-lasting (tonic) effects on CCs. A new realm of single-trial ERP quantification is possible in that the MC appears as a marker of a single cognitive variable whereas the CC appears as a marker of a series of functionally related cognitive variables. The engagement of the brain in a specific cognitive task is accompanied by an increase of CCs in single-trial ERPs, which is especially informative with respect to the P3 cognitive potential. New methodology provides evidence for the three basic types of single-trial P3 sub-components: monolithic P3a, monolithic P3b, and a complex component, P3ab, which includes both P3a and P3b as sub-components.