Band power features correlate with performance in auditory brain-computer interface
Machine Learning Group, Berlin Inst. of Tech., Germany
Brain-Computer Interfaces (BCI) aim at providing a means of communication and control, mediated via the detection and decoding of specific brain states. In order to reach practical applicability, the classifiers of brain activity employed in a BCI must work reliably and robustly over extensive periods of time. In this contribution, we investigate fluctuations of classifier performance that occurred during the online feedback phase of a BCI spelling application (Hoehne et al. 2010), which was based on the auditory evoked potential (AEP). For this purpose, the experimental data of 10 participating subjects is re-analyzed offline. First a continuous measure for classifier performance is defined. The measure is based on classification rates of individual stimulus presentations. Then the linear interactions between the performance measure and band power features of various frequency bands are determined. We find positive correlations between the performance measure and parietal-occipital alpha power (8 to 14 Hz). Furthermore, we observe negative correlations between the performance measure and the power of central (centro-parietal) low (high) range gamma oscillations. The mentioned findings reach significance on the group level analysis as well as in the majority of individual subjects. The observed negative correlations in central and parietal gamma oscillations are in line with recent findings on the relation between gamma power and performance in an entirely different BCI paradigm, namely motor imagery (Grosse-Wentrup et al., 2010). Thus, our results stress the importance of neural oscillations for information processing in the brain. Funding: This work is supported by the European ICT Programme Project FP7-224631.
XI International Conference on Cognitive Neuroscience (ICON XI), Palma, Mallorca, Spain, 25 Sep - 29 Sep, 2011.
Poster Sessions: Neuropsychiatric Applications
(2011). Band power features correlate with performance in auditory brain-computer interface.
Front. Hum. Neurosci.
XI International Conference on Cognitive Neuroscience (ICON XI).
16 Nov 2011;
25 Nov 2011.
Dr. Sven Dähne, Machine Learning Group, Berlin Inst. of Tech., Berlin, Germany, firstname.lastname@example.org