Biomedical Engineering Reference
In-Depth Information
Fig. 5.
Electrode placement for the 10-10 montage
estimation method [18]: the realistic sources generated as described in the previous
subsection were assumed to be time courses of brain dipoles. The positions and the
orientations of these dipoles (8 to 48) were randomly generated inside the inner sphere
of a the head model, representing the brain. The six parameters of the dipoles being thus
completely defined (Cartesian coordinates, angles and magnitude), one can generate the
corresponding scalp map (i.e. the electrical potentials on the outer sphere) using the so-
called Lead Field matrix corresponding to the three-spheres forward model (see for
example [22]). We used here the Berg technique [18], which accurately and rapidly
approximates the scalp potentials generated by a dipole for a three spheres model with
the weighted sum of the potentials generated by three dipoles in a single sphere model.
The Lead Field matrix allows the computation of the scalp map (potentials) for ev-
ery point on the surface of the outer sphere. Still, in the EEG simulation context, we
are only interested by the potentials recorder by the scalp electrodes. In our simu-
lation, we used as a basis montage the classical 10-10 EEG montage (figure 5). We
have next chosen five subsets consisting of 8, 16, 24, 32 and 48 electrodes correspond-
ing to real EEG applications (sleep studies, brain computer interfaces and clinical se-
tups for epilepsy diagnostics).For example, for the 8-electrodes montage, the chosen
electrodes were
F
pz
,T
7
,C
3
,C
z
,C
4
,T
8
,O
1
and
O
2
; for the 16-electrodes montage,
F
3
,F
z
,F
4
,T
7
,C
3
,C
z
,C
4
,T
8
,P
7
,P
3
,P
z
,P
4
,P
8
,PO
7
,PO
8
,O
z
; and son on.
5
Results and Discussion
The four algorithms are first evaluated on a random data set in order to define a min-
imum data length rule ensuring accurate separation performances and to analyse the
impact of initialization step in the general toy case of unstructured randomly mixed
data. An equivalent study is then applied on a more physiologically plausible data set,
on which our minimum data length rule is validated. On the same plausible data, the re-
sults are evaluated in order to confirm or infirm the superiority of sphering initialization
Search WWH ::
Custom Search