From Neural Network Code to Brain/Machine Interface
27-29 September - 2007 - Gouvieux les Chantilly (France)
 
Neurofeedback for modeling mechanisms of biological control of EEG rhythms
 
Arnaud Delorme (1,2), Julie Onton(2), Paul Hammon(2), Scott Makeig(2)
 

(1) Centre de Recherche Cerveau & Cognition, Toulouse, France

(2) Swartz Center for Computational Neuroscience, UCSD, La Jolla, USA

arno@salk.edu

 
    'Closing the [brain] loop' by providing near-immediate sensory feedback about one's own EEG rhythms has long been feasible, and was popularly advocated by some 30 years ago for relaxation and altering mental state, Today, research, experimentation, peri-professional practice in this area is again increasing in at least three directions. First, the field of brain-machine interface (BMI), or brain-computer interface (BCI), is attracting widespread interest from both rehabilitation specialists and engineers. Second, a growing number of unlicensed practitioners of 'neurofeedback' therapies claim to be able to treat a wide range of physical, mental, emotional and behavioral conditions, though with little formal medical research to back up their claims and observations and less understanding of the biological mechanisms involved. Finally, there are continuing efforts to use EEG feedback for mental state monitoring in transportation and other work places. Proposals range from monitoring the emergence of subject drowsiness during long pilot work shifts to near-instantly reading out early target-related brain responses during rapid image presentation.
    
    We are currently investigating the application of independent component analysis (ICA) to the design of BCI, mental monitoring, and/or neurofeedback interfaces. While most current studies involve the feedback of frequency information from one or two EEG channels, we propose a system where feedback delivers time/frequency information about independent components of the EEG. The signal at a single electrode represents the sum of many different sources of electric brain activity. Therefore, to control the signal recorded at a single electrode, one may have to control the activities of multiple cortical sources. In contrast, an independent component accounts for the contribution of an isolated cortical source to the entire set of scalp electrodes.
    
    We are particularly interested in testing whether such feedback may improve overall task performance, in particular during expert monitoring and learning. Such improvement might be expected if the independent component process variables controlled are involved in or, more broadly, are controlled by brain systems that regulate cognitive processes involved in optimum performance, for example attention and working memory. We have built a simple neurofeedback paradigm, based on the Wolpaw-lab BCI2000 software, to test this hypothesis, and are currently running pilot experiments to refine our experimental procedures. We believe that these experiments can also address fundamental questions about control of cortical EEG. Is it possible to learn to control a single independent EEG component without affecting the activities of other components?