Classifying Single Trial EEG:Towards Brain Computer Interfacing
, Gabriel Curio
and Klaus-Robert Müller
Fraunhofer-FIRST.IDA, Kekuléstr. 7, 12489 Berlin, Germany
Neurophysics Group, Dept. of Neurology, Klinikum Benjamin Franklin,Freie Universität Berlin, Hindenburgdamm30, 12203 Berlin, Germany
University of Potsdam, Am Neuen Palais 10, 14469 Potsdam, Germany
Drivenbythe progressin theﬁeld ofsingle-trialanalysis of EEG,there isa growing interest in brain computer interfaces (BCIs), i.e., systems thatenablehumansubjects to controla computeronlybymeans oftheir brainsignals. In a pseudo-online simulation our BCI detects upcoming ﬁngermovements in a natural keyboard typing condition and predicts their lat-erality. This can be done on average 100–230ms
the respectivekey is actually pressed, i.e., long before the onset of EMG. Our approachis appealing for its short response time and high classiﬁcation accuracy(
96%) in a binary decision where no human training is involved. Wecomparediscriminativeclassiﬁers like SupportVectorMachines (SVMs)and different variants of Fisher Discriminant that possess favorable reg-ularization properties for dealing with high noise cases (inter-trial vari-ablity).
The online analysis of single-trial electroencephalogram (EEG) measurements is a chal-lenge for signal processing and machine learning. Once the high inter-trial variability (seeFigure 1) of this complex multivariate signal can be reliably processed, the next logicalstep is to make use of the brain activities for real-time control of, e.g., a computer. In thiswork we study a
evaluation of single-trial EEGs from voluntary self-pacedﬁnger movements and exploit the laterality of the left/right hand signal as one bit of infor-mation for later control. Features of our BCI approach are (a)
for artifacttrials, (b) state-of-the-artlearning machines with inbuilt feature selection mechanisms (i.e.,sparse Fisher Discriminant Analysis and SVMs) that lead to
96% classiﬁcation accura-cies, (c)
and (d) short response times. Although our setup was not tunedfor speed, the
determined information transmission rate is 23 bits/min whichmakes our approach competitive to existing ones (e.g., [1, 2, 3, 4, 5, 6, 7]) that will bediscussed in section 2.
To whom correspondence should be addressed.