Information Technology Reference
In-Depth Information
component and the ERS component. Indeed, our classi
cation results showed that
the theta ERS brings independent
cation
results (as compared with using the ERP alone). It has to be noted that while the Ne
was clearly identi
information and allows better classi
able in all subjects, the Pe was not strong enough to be clearly
identi
ed in some subjects. This might explain why our BSS approach has not been
successful in
finding separated sources for the Pe peaks (poor signal-to-noise ratio
and/or high interindividual variability).
Interestingly, we have found that the expectation of the outcome has a direct
impact on the theta ERS, but not on the Ne; the more the error is expected, the
weaker is the theta ERS. To our knowledge, no such effect has been reported so far.
We conclude that the error-related potential may depend on two factors: the value
of the observation (erroneous or correct) and the expectation of the outcome. Thus,
the error-related potential may be the combination of two reactions, one to the error
and the other to the surprising character of the observation. Further studies may
now try to investigate this new aspect of the error-related potential and try to
determine whether these two components are physiologically separated or inter-
laced. Within the frame of a BCI application, the more accurate the BCI is, the more
unexpected the error will be. Classi
cation results showed that when using theta
component, performance is higher for unexpected errors as compared to expected
errors. If the subject is concentrated and performs well the task, the occurrence of an
error will be less expected, since it would result mainly from a nuisance such as an
artifact decreasing the signal-to-noise ratio. Under these circumstances, the theta
ERS component will be more ef
cient in detecting errors coming directly from the
interface. In order to improve ErrP recognition in a real BCI system, the perfor-
mance of the system should be maximized, so that the ErrP can be more easily
detected. We conclude that the theta ERS will be stronger for high-performance
BCIs and therefore that the error can be more easily detected for high-performance
BCI. This fact should be taken into consideration in ensuing attempts to integrate a
control loop based on ErrP detection in a BCI. More in general, the error potential
should not be seen as a panacea for correcting BCI operation errors, since a high
number of errors will lead to a poor detection of ErrP.
In conclusion, the AJDC method proves at the same time
flexible and powerful.
We hope that it turns out useful for extracting meaningful information to be used in
the studies at the crossroad of music and brain electrophysiology.
8.11 Questions
1. What are the physical generators of brain electric
fields recordable from the
scalp?
2. Why a linear mixing model
is a good approximation for the genesis of
observable scalp potentials?
3. What is the relation between the mixing matrix and the demixing matrix?
