Information Technology Reference
In-Depth Information
Then, in this study, the feedback corresponds to the actual performance achieved in
the task, again approximating the actual operation of a BCI. Finally, the memory
task continuously adapts to the ability of the participants during the whole exper-
iment. This ensures that the cognitive load is approximately constant across the
duration of the experiment, that it is comparable across individuals regardless of
their memory span, and that the error rate across subjects is approximately equal.
This latter point is particularly important in ErrP studies since it is known that the
error rate affects the ErrP ([8]). In this study, the adaptive algorithm is tuned to
engender an error rate of about 20 %, which amount approximately to the rea-
sonable accuracy of a reactive BCI operation in real-world situations.
(2) New Multivariate Signal processing Analysis
cation (correct vs. error)
have reached encouraging results (around 70 % of overall accuracy) using only
little a priori knowledge on this potential. As usual, a more profound knowledge of
the electrophysiological characteristics of the ErrPf can be used to select more
relevant and robust features for the purpose of single-trial online detection. Previous
studies showed that the ErrP can be characterized in the temporal domain both as an
ERP (time- and phase-locked event) and as an event-related synchronization, or
ERS (time- but non-phase-locked event). The ERP is characterized by a negative
de
Some of the previous studies on single-trial ErrP classi
ection, named Ne, sometimes followed by a positive one named Pe (Gentsch
et al. 2009 ; Steinhauser and Kiesel 2011 ). The ERS is characterized by an increased
oscillatory activity in the theta frequency band-pass region (4
7.5 Hz) occurring
approximately in the same time window and spatial location as the Ne (Trujillo and
Allen 2007 ). Source localization of the FRN using dipole analysis has suggested
generators in the anterior cingulate cortex (ACC) and the supplementary motor area
(Gehring and Willoughby 2002 ; Miltner et al. 1997 ). Similar results have been
obtained for the ErrPr. Hereby, we propose a sharp spatial
-
filtering approach based
on the blind source separation approach described above with the aim to disen-
tangling the sources responsible for the ERP and the ERS; if this proves feasible,
then the ERP and ERS components will yield independent features to feed the
classi
er, hence potentially increasing the online accuracy.
As a
first objective, we identify the different components of the ErrP along
dimensions time, space, and frequency by means of a multivariate analysis both in
the sensor space and in the source space. We jointly estimate the brain sources at
the origin of the ERP and ERS components and assess their different roles in error
reaction. Then, we study the role of these components on the ErrP with respect to
the expectation of participants. Finally, we look at how these results impact on ErrP
single-trial classi
cation, which is the essential step in integrating ErrPs in BCI
systems.
 
Search WWH ::




Custom Search