Biomedical Engineering Reference
In-Depth Information
0.5
0
0.5
0.5
0
0.5
0.5
0
0.5
0.1
0
0.1
0
200
400
600
800
1000
time points
Fig. 5.2
The time courses of the three sources are shown in the
upper
three panels. The
bottom
panel
shows the signal-only magnetic recordings (before noise is added). The selected three time
courses from sensor #1, #101, and #201 are shown. The ordinates show relative values, and the
abscissa shows the time points. A total of 1,200 time point data was generated
Simulated magnetic recordings were computed by projecting the time courses of the
three sources onto the sensor space using the sensor lead field at the source locations.
These recordings are shown for three selected sensor channels in the bottom panel
of Fig.
5.2
.
We first performed denoising experiments using the BFA and VBFA algorithms.
We generated simulated MEG data by adding the sensor noise onto the computed
magnetic recordings shown in the bottom panel of Fig.
5.2
. Here, the noise generation
was performed using a Gaussian random number, and the signal-to-noise ratio (SNR)
was set to one. The noise-added simulated sensor data is shown in the top panel of
Fig.
5.3
for the same selected sensor channels.
The BFA algorithm was applied to these simulated MEG data and the signal
component was estimated by computing
A
u
k
. The results of this experiment are
presented in Fig.
5.3
. The results with the number of factors
L
set to 3 are shown in
the second panel from the top, and the results with
L
set to 20 are in the third panel.
Since we assume that three sources exist in this numerical experiment, the correct
value of
L
is three.
¯
