Comparison of single trial back-projected independent components with the averaged waveform for the extraction of biomarkers of auditory P300 EPs

Abstract

The independent components analysis (ICA) of the auditory P300 evoked responses in the EEG of normal subjects is described. The purpose was to identify any features which might provide the basis for biomarkers for diseases, such as Alzheimer’s disease. Single trial P300s were analysed by ICA, the activations were back-projected to scalp electrodes, many artefactual components were removed automatically, and the back-projected independent components (BICs) were first clustered according to their amplitudes and latencies. Then these primary clusters were secondarily clustered according to the columns of their mixing matrices, which clusters together those BICs with the same scalp topographies and, therefore, source locations. The BICs comprising the P300s had simple shapes, approximating half-sinusoids. Trial- to-trial variations in the BICs were found, which explain why different averages have been reported. Both positive- and also negative-going BICs were identified, some associated with known peaks in the P300 waveform. Artefact-free, single trial P300 waveforms could be constructed from the BICs, but these are probably of less interest than the BICs themselves. The findings demonstrate that neither averaged P300s, nor single trial P300s, are reliable as biomarkers, but rather it will be necessary to investigate the BICs present in a number of single trial realizations.