EN FR
EN FR


Section: New Results

Brain Computer Interfaces

Multimodal BCI

Participants : Maureen Clerc, Lorraine Perronnet [Visages project-team] , Saugat Bhattacharyya [Camin team] .

We are conducting research in Multimodal BCI:

  • In collaboration with Camin team in Montpellier, we are investigating the use of feedback using Functional electrical stimulation (FES) of limb muscles  [58] for Motor Imagery and also studying the influence of the FES on brain signals.

  • A study comparing unimodal and bimodal EEG-fMRI neurofeedback for 10 healthy volunteers showed that EEG-fMRI leads to stronger activations than EEG alone [24].

Automatizing calibration

Participants : Maureen Clerc, Nathalie Gayraud, Alain Rakotomamonjy [Université de Rouen] .

One of the drawbacks of BCI is the time required for setup and calibration before its use. Instead of fine-tuning the BCI by collecting labeled data by asking the user to perform tasks without any purpose nor feedback, we propose to fine-tune the BCI after the user has started using it. This requires an initial - suboptimal - classifier, which we propose to build through “transfer learning” by re-using labeled data acquired from other subjects and other sessions. We have investigated two main directions for this:

  • Riemannian geometry of covariance matrices. Covariance matrices of EEG signals are interesting features for BCI. Their information geometry has led to impressive transfer learning performance, as testified by their excellent ranking in several competitions. We are studying the advantages of these features and how they can be used to build separability markers within datasets  [74].

  • Optimal transport theory. A new strand of research is to use optimal transport methods for domain adaptation. The idea is to reuse the classifiers built from existing labeled datasets by transporting the new unlabeled data onto the domain of the existing data [34].

Translational research

Participants : Maureen Clerc, Claude Desnuelle [CHU de Nice] , Violaine Guy [CHU de Nice] , Théodore Papadopoulo, Marie-Hélène Soriani [CHU de Nice] .

The P300-speller is a widespread BCI paradigm for communication, studied in many laboratories. Our involvement in this paradigm was triggered by the Nice University Hospital ALS reference center. Having evaluated with them existing P300-spellers, which were found difficult to get to work properly, we decided to develop our own P300-speller based on OpenViBE in collaboration with Inserm Lyon  [65], [104]. Among its distinctive features: optimal stopping of flashes, principled choice of letter groups  [103] and word completion and prediction. We demonstrated the feasibility of our “Coadapt P300 speller” in collaboration with Nice University Hospital during a clinical study with 20 ALS patients who participated in 3 sessions each  [99][20].

In order to bring this type of communication BCI closer to patients, we developed a user-friendly software, bci-vizapp, with far greater portability with respect to hardware (OS, screen and amplifier).

Our work aroused the interest of patient associations, in particular “Espoir Charcot” who helped a patient hospitalized in Chambéry acquire a consumer-grade (Emotiv-EPOC) EEG in order to use the P300-speller. He eventually succeeded in using the system, with the help of a local engineer, but notably without our physical presence at any stage. This represents an important first step for us in translational research.