On the Independent Component Analysis of evoked potentials through single or few recording channels

In this work we propose a technique based on independent component analysis (ICA), applied to single or two channel(s) recordings of electroencephalogram (EEG) brain signals. Standard (ensemble) ICA requires multiple channel recordings to work, however when single of few channels are required ensemble ICA cannot be readily applied. Single channel ICA (temporal ICA) can be performed by preprocessed the data using the method of delays. Few channels (space-time ICA) can be analysed in an extension to this method. These techniques are demonstrated on the P300 evoked potentials (EPs) of a brain-computer interfacing (BCI) word speller dataset. We furthermore show how it is possible to extract single trial evoked EPs (i.e. non-stimulus locked) within a little as 3 epochs and even on channels not over the event focus. Due to the poor SNR, as well as the presence of other artifacts, it is difficult to detect the P300 pattern on raw signal data. The results show that proposed algorithms are able to accurately and repeatedly extract the relevant information buried within noisy signals and to do so without the requirement of stimulus locked averages. These advantages are paramount for building a more reliable and robust system for use in real-world BCI - i.e. for use outside of the clinical laboratory.