Biomedical Engineering Reference
In-Depth Information
0.2
0.1
0
0.1
0.2
0.1
0
0.1
0.1
0
0.1
0.1
0
0.1
0
200
400
600
800
1000
time points
Fig. 5.3
Results of denoising experiments using BFA and VBFA. The
top panel
shows the sensor
recordings before denoising. The
second panel
from the
top
shows the denoising results by BFA
with the model order
L
set to three. The
third panel
shows the denoising results by BFA with
L
set
to twenty. The
bottom panel
shows the denoising results by VBFA with
L
set to twenty. The time
courses of the selected three sensor channels, (channel #1, channel #101, and channel #201), are
shown. The ordinates show relative values, and abscissa shows the time points
Results in the second panel (results with the correct
L
) are better than the results in
the third panel (results with a significantly overespecified
L
), and we can see that the
overspecification of the number of factors,
L
, reduces the denoising capability of the
algorithm. The VBFA algorithm was then applied, and the results are shown in the
bottom panel of Fig.
5.3
. Here, although the number of factors
L
was set to 20, results
close to those obtained using the BFA algorithm with the correct value of
L
were
obtained. The results demonstrate the fact that the VBFA algorithm incorporates the
model order determination and is insensitive to the overspecification of the number
of factors.
We next performed experiments on the interference removal using the PFA algo-
rithm. In this experiment, the simulated MEG recordings were generated with 2,400
time points where the first half period does not contain the signal activity, which
is contained in the second half period. The interference data was computed by
