An ECoG Based Brain Computer Interface with Spatially Adapted Time-Frequency Patterns

In this paper we describe an adaptive approach for the classification of multichannel electrocorticogram (ECoG) recordings for a Brain Computer Interface. In particular the proposed approach implements a timefrequency plane feature extraction strategy from multichannel ECoG signals by using a dual-tree undecimated wavelet packet transform. The dual-tree undecimated wavelet packet transform generates a redundant feature dictionary with different time-frequency resolutions. Rather than evaluating the individual discrimination performance of each electrode or candidate feature, the proposed approach implements a wrapper strategy to select a subset of features from the redundant structured dictionary by evaluating the classification performance of their combination. This enables the algorithm to optimally select the most informative features coming from different cortical areas and/or time frequency locations. We show experimental classification results on the ECoG data set of BCI competition 2005. The proposed approach achieved a classification accuracy of 93% by using only three features.