Wang, S. and James, C.J.
Enhancing evoked responses for BCI through advanced ICA techniques
In Proceedings of the 3rd International Conference on Advances in Medical, Signal and Information Processing (MEDSIP 2006).
Institution Of Engineering And Technology..
Full text not available from this repository.
An electroencephalogram (EEG) based brain-computer interface (BCI) is a communication system in which messages or commands that an individual sends to the external world do not pass through the brain’s normal output pathways but is detected through EEG activity. The overall goal is to provide those users with severe mobility disability basic communication capabilities to interact with their environment. The P300 word speller is one of the important BCI applications which detects real-time evoked brain signals and translates them into letters (and then words) within a particular BCI paradigm. However due to the poor SNR of EEG as well as the presence of other artifacts, the identification accuracy is still not high enough for real-world application. Here we present three slightly different approaches to improving performance based on ICA including: standard ICA, ICA with template assisted component selection and spatially-constrained ICA. To evaluate the ICA performance, we only consider a very simple linear detector in the classification section. When compared with the classification results obtained from using the raw unprocessed data, the results using these approaches show distinct improvement: a maximum accuracy of 96.8% versus 51.6% maximally from raw data. Furthermore, the results after ICA indicate that it is possible to reduce the number of repeated epochs required to perform stimulus locked averages, whilst still maintaining good performance measures. This has the potential of speeding up the word speller and has further implications for use on similar ERP based systems.
Actions (login required)