Finding Significant Correlates of Conscious Activity in Rhythmic EEG
- PDF / 847,448 Bytes
- 6 Pages / 600 x 792 pts Page_size
- 4 Downloads / 200 Views
Finding Significant Correlates of Conscious Activity in Rhythmic EEG Piotr J. Durka Laboratory of Medical Physics, Institute of Experimental Physics, Warsaw University, ul. Ho˙za 69, 00-681 Warsaw, Poland Email: [email protected] Received 28 January 2004; Revised 27 July 2004 One of the important issues in designing an EEG-based brain-computer interface is an exact delineation of the rhythms, related to the intended or performed action. Traditionally, related bands were found by trial and error procedures seeking maximum reactivity. Even then, large values of ERD/ERS did not imply the statistical significance of the results. This paper presents complete methodology, allowing for a high-resolution presentation of the whole time-frequency picture of event-related changes in the energy density of signals, revealing the microstructure of rhythms, and determination of the time-frequency regions of energy changes, which are related to the intentions in a statistically significant way. Keywords and phrases: time-frequency, adaptive approximations, matching pursuit, ERD, ERS, multiple comparisons.
1.
INTRODUCTION
Thinking of a “brain-computer interface” (BCI), one can imagine a device which would directly process all the brains output—like in a perfect virtual reality machine [1]. Today’s attempts are much more humble: we are basically at the level of controlling simple left/right motions. On the other hand, these approaches are more ambitious than direct connections to the peripheral nerves: we are trying to guess the intention of an action directly from the activity of the brains cortex, recorded from the scalp (EEG). Contemporary EEG-based BCI systems are based upon various phenomena like, for example, visual or P300 evoked potentials, slow cortical potentials, or sensorimotor cortex rhythms [2]. The most attractive path leads towards the detection of the “natural” EEG features, for example a normal intention of moving the right hand (or rather its reflection in EEG) would move the cursor to the right. Determination of such features in EEG is more difficult than using evoked or especially trained responses. Desynchronization of the µ rhythm is an example of a feature correlated not only with the actual movement, but also with its mere imagination. All these approaches encounter obstacles, common in the neurosciences: great intersubject variability and poor understanding of the underlying processes. Significant improvement can be brought by coherent basic research on the EEG representation of conscious actions. This paper presents two methodological aspects of such research. (i) High-resolution parameterization and feature extraction from the EEG time series. Scalp electrodes gather
signal from many neural populations, so the rhythms of interest are buried in a strong background. Owing to the high temporal resolution of EEG and the oscillatory character of most of its features, we can look for the relevant activities in the time-frequency plane. (ii) Determination of significant correlates of conscious activities requires
Data Loading...
