Bipolar filters improve usability of Brain-Computer Interface technology in post-stroke motor rehabilitation

The development of usable and accurate brain-computer interface (BCI) systems enables the transfer of this technology to clinical routine. When working with electroencephalographic signals (EEG), an important factor to optimize the signal to noise ratio of the signal is to choose the appropriate spatial filters. Specific aims of this study were (a) to compare classification performances of two commonly used filters, two bipolar filters (longitudinal and transversal) and the combination of both bipolar filters obtained by pooling EEG features together, (b) to compare the number of physical electrodes needed as consequence of the spatial filter choice. Bipolar filters showed classification performances comparable to those provided by the most commonly used filters, despite requiring a significantly lower number of electrodes. Longitudinal bipolar filters showed the best accuracy to number of electrodes ratio; thus, its usage is suggested for applications such as upper limb motor rehabilitation in post-stroke patients