TOPICS 

Our research focus falls within the definition of Computational Neuroscience, i.e., an interdisciplinary field which draws on applied mathematics, physics and computer science to understand, describe and predict the nervous system and its pathologies. At the moment our research projects can be framed within three major lines:

1) Understanding the neural mechanisms used by neural populations to encode/decode sensory-motor information: We are currently using pattern recognition techniques to understand the role played by the different oscillations within the neural code. The use of pattern recognition allows understanding and mimicking the coding/decoding processes that are carried out by neural populations in a trial-by-trial basis.

2) Development and evaluation of techniques to non-invasively study the brain electromagnetic activity in healthy subjects and patients: A traditional research topic of the members of this group has been the design, evaluation and application of different inverse solutions. One important aspect of the new research lines is the development of robust inverse solution methods for the analysis of single trials rather than averages over stimuli repetitions.

3) Bayesian modeling of perception and action: How the brain deals with noise and uncertainty: To use sensory information efficiently to make judgments and guide action, the brain must represent and use information about uncertainty in its computations for perception and action. This leads to the Bayesian coding hypothesis: that the brain represents sensory information probabilistically, in the form of probability distributions.One of our aims is to test the Bayesian coding hypothesis experimentally, and so determine whether and how neurons code information about sensory uncertainty.

Back to TOP

                                               COLLABORATIONS


Prof. Katalin Gothard

Prof. Olaf Hauk

Stephen Perrig, M.D.   Laboratoire du Sommeil. Neuropsychiatrie. HUG

Carles Grau Fonollosa. Department of Psychiatry and Clinical Psicobiology, University of Barcelona







Back to TOP


                                                    PROJECTS

-European Project MAIA FP6-3758 MAIA

  The contributions of the Electrical Neuroimaging Group  to MAIA project includes:

1- Use of local field potentials non invasively estimated with ELECTRA for the development of BCI and the comparison of their information content with EEG and real invasive intracranial recordings.

2- Inclusion and use of the high frequency oscillatory activity of the EEG as the basis of physiologically based features.

3- Proposal of efficient methods (Discriminative Power) for the identification and selection of physiological features.

4- Proposal and evaluation of BCIs based on covert and overt visual attention using Posner like paradigms  as well as steady state visual evoked potentials (SSVEP).

The main objective of MAIA was "to develop a non-invasive direct brain-computer interface (BCI) that determines the subject’s voluntary intent to do a large set of primitive motor actions on the order of milliseconds and conveys this intention to a robot that will implement the necessary low-leveldetails for achieving complex tasks".

To achieve that goal we have proposed the use of the SSVEP BCI with the following properties:

1) As the motor intention, the "sight" precedes naturally several movements of the body. In that sense SSVEP is closer to motor intention than motor imagery.

2) Allows for a perfect cassification (100%) of several simultaneous classes using very short time periods. For the limited needs of MAIA (control of a wheelchair) we proposed a system able to identify 4 classes using EEG windows of 0.25 to 0.5 seconds.

3) In practice this BCI controlled of  a robot simulator (see download) and a real robot via internet without any artificial intelligence.

It has been erroneously suggested that SSVEP performance is due to foveating. In fact SSVEP are based on a property of some primary sensorial brain regions that "enter in resonance" with the frequency of an external stimulus. The intensity of this response is modulated by the subject attention. This property allows for very short time windows in contrast to motor imagery, word association, and other (unnatural) methods that need several seconds to change from one state to the other.

For a comparison with motor imagery BCI see download page.

A word of caution. Several demonstrations use the BCI shared system, that is,  a combination of the BCI and the artificial intelligence of the robot. Under these conditions nothing can be said about the BCI until it is not tested alone, i.e., without  the obtacle avoidance strategies of the robot. For details see The principle of shared autonomy and the evaluation of BCIs.

  Main Conclusions from the  FP6-3758 MAIA project:

The reviewers recognized it an "acceptable project" with "very good research results" and considered that "the research performed on very high frequency oscillations (VHFO) revealed interesting aspects which are of fundamental interest for a better understanding of neural processing."

As for a criticism they remarked  that "some of the initial goals of the project were not achieved."


From our side we have  identified the following mean weaknesses:

- The BCI system used on the public demonstrations (by IDIAP and KUL)  does not satisfy any of the initial goals of MAIA about the identification of more classes in less time and are not based on the recognition of subjects intent. Instead of that, the system demonstrated by IDIAP uses complex and unnatural mental tasks (motor imagery, word association and relax state) using a (visual or muscle) artifact to stop the BCI.  

-The fact that the computer is sending commands every 0.5 seconds does not mean that the subject can produce different and identifiable mental states in the same time period.

- The movies describing IDIAP BCI system have received systematic criticism during public presentations and are considered more as a demonstration of the robot intelligence than the result of an efficient BCI. That is, several researchers consider that the shared control is masking the real behavior of the BCI.

- From the configurations files (distributed to all MAIA partners) it is clear that the simulator used by IDIAP contains especial agents ("Center of Corridor" and "Docking" that may be active even if the intelligence is set to NONE) which artificially correct or keep the trajectory of the robot and thus prevent the real control of the robot by the subject.

- Demontrations using the robot simulator does not include the graphical element to identify the intelligence level used or  the possible active agents (e.g. "stop before collision" "obtacle avoidance", etc). We would note that the only two demonstrations using the IDIAP BCI alone (i.e. without intelligence) have finished by a failure (see download) .  


-European Project BACS


-National Project: IM2-BMI

  The IM2 white paper (2002) was one of the first Swiss National projects on BCI. On this framework we proposed for the first time (at both national and international level) the use of inverse solutions as the basis of direct non invasive brain computer interfaces as well as the identification of physiologically meaningful features based on the current knowledge about brain functioning. 

-Generalitat de Catalunya. Grup de Recerca Consolidat.

-Generalitat de Catalunya. Xarxa Temàtica