NIPS 2000 Workshops
Breckenridge, CO

NIPS2000 Workshop on
"Algorithms, technologies and neural representations
for neuroprosthetics and neurorobotics
."

Steps toward EEG-based prosthetic control

Scott Makeig, Tzyy-Ping Jung and Terrence J Sejnowski
Institute for Neural Computation,
University of California San Diego, La Jolla CA.
and The Salk Institute, La Jolla CA 92037
scott@salk.edu

To build a neuroprosthetic based on voluntary control of EEG signals, several steps are necessary. First, the spatiotemporal information in spontaneous and/or evoked EEG data must be separated from the data. For this purpose, independent component analysis (ICA) is very promising. The event-related spectral perturbations (ERSPs) of independent components of the EEG show clear linkages between cognitive and motor events and changes in the amplitudes and phase dynamics of synchronous brain oscillations at multiple EEG frequencies. For example, we have found that the suppression of the mu rhythm generated in and near the motor cortex prior to an anticipated hand movement (currently used by some groups to effect prosthetic control) is 4-6 dB larger after ICA separation than in any single channel derivation (Makeig et al., IEEE Trans Rehab Eng 8:208-11, 2000. (.pdf, 111k)). Next, the degrees of learned voluntary control over each of the independent EEG components must be evaluated as well as the degrees of freedom this control can achieve. With this knowledge, suitable robotic prostheses may be designed to take advantage of the learned control for useful purposes by paralyzed individuals. Other applications include predicting the responsiveness of operators of complex systems to imperative stimuli (Jung et al., 1996).

Return to Current Abstracts


NIPS2000 Workshop on
"Explorative analysis and data modelling in functional neuroimaging:
Arriving at, not starting with a hypothesis!
"

Should we expect convergence between concurrent EEG and BOLD data?

Scott Makeig
Computational Neurobiology Lab, Salk Institute, La Jolla, CA
Institute for Neural Computation, University of California San Diego, La Jolla CA
scott@salk.edu

I will consider the basis for expecting conjunction or disjunction between the results of concurrent fMRI and EEG neuroimaging, and will illustrate the application of ICA to continuous EEG data decomposition. While much effort has been expended to find common areas of activation for (a) particular EEG features of event-related potential (ERP) averages and for (b) areas active in fMRI experiments, recording concurrent EEG and fMRI data provide a much richer basis for exploration of what kind of "neural" activity is correlated with fMRI BOLD signal changes. I will give a simple argument from first principles suggesting there need be no linkage at all between brain sources of ERP and EEG features, on one hand, and fMRI BOLD signal changes on the other. Concurrent EEG and fMRI data collection and analysis presents several challenges. I will illustrate the usefulness of ICA combined with single-trial visualization methods to recover event-related EEG brain dynamics during continuous high-rate fMRI data acquisition.

Return to Current Abstracts