Single-trial detection of somatosensory evoked potentials by probabilistic independent component analysis and wavelet filtering

ObjectiveTo develop an effective approach for enhancing the signal-to-noise ratio (SNR) and identifying single-trial short-latency somatosensory evoked potentials (SEPs) from multi-channel electroencephalography (EEG).Methods128-channel SEPs elicited by electrical stimuli of the left posterior tibial nerve were recorded from 11 healthy subjects. Probabilistic independent component analysis (PICA) was used as a spatial filter to isolate SEP-related independent components (ICs), and wavelet filtering was used as a time–frequency filter to further enhance the SNR of single-trial SEPs.ResultsSEP-related ICs, identified using PICA, showed typical patterns of cortical SEP complex (P39–N50–P60) and scalp topography (centrally distributed with the spatial peak located near vertex). In addition, wavelet filtering significantly enhanced the SNR of single-trial SEPs (p = 0.001).ConclusionsCombining PICA and wavelet filtering offers a space–time–frequency filter that can be used to enhance the SNR of single-trial SEPs greatly, thus providing a reliable estimation of single-trial SEPs.SignificanceThis method can be used to detect single-trial SEPs and other types of evoked potentials (EPs) in various sensory modalities, thus facilitating the exploration of single-trial dynamics between EPs, behavioural variables (e.g., intensity of perception), as well as abnormalities in intraoperative neurophysiological monitoring.

Research highlights
► Spatial filter based on PICA to isolate SEP-related ICs and enhance the SNR of ERPs.
► Time–frequency filter based on continuous wavelet filtering to significantly enhance the SNR of ERPs.
► Combining PICA and wavelet filtering offers a space–time–frequency filter to provide a reliable estimation of single-trial ERPs.