Biomedical Engineering Reference
In-Depth Information
frequency. Similar issues can be seen with the 40 Hz response, which is mainly due to
the decrease in the SSVEP strength at the very high frequency.
As expected, the overall performance of the water-based electrodes was much higher
than for the dry electrodes. The surprising result was that the overall accuracy obtained
with gel electrodes was comparable to the one obtained with water-based ones. Al-
though this coincides with the results obtained in [41], as we did not used any advanced
algorithm for SSVEP response estimation, these results were unexpected. To better un-
derstand these results we analyzed the impact of noise on the signal, and the signal
quality in different electrodes.
Spectral Content across Electrode Types and Positions. By comparing the raw sig-
nal and the power spectra obtained within different EEG channels we observed the
following:
1. The noise component in the EEG signal, being environmental or physiological, can
be observed for all setups.
2. The severity of noise contribution in the signal and the number of EEG channels
contaminated by the noise is higher for the dry setup than the water-based setup,
and for the water-based than the gel setup.
3. The impact of noise per electrode can vary throughout a single recording session
and it differs for different recording sessions.
4. In most cases, the higher the level of noise in an EEG channel, the lower the SSVEP
response.
The first observation can be explained by the environment which was not specially
shielded for such kind of experiment and could have been contaminated by electromag-
netic waves. Also, motion artifacts stemming from muscle tension and head and body
movements were present, due to the lengthy recording procedure. The second observa-
tion was expected due to the type of the skin-electrode contact. The third one was not
expected for gel electrodes, although it can be partially explained by the fast preparation
procedure that did not include skin cleaning and de-greasing before the measurement.
It was expected for water-based and especially dry electrodes. Finally, the last obser-
vation was a learning point for us and we wanted to incorporate this fact in devising
the electrode selection algorithm that will improve the accuracy levels obtained with
different setups, as explained below.
4.2
The Impact of Noise Estimation on the Electrode Selection
To estimate the noise level in the signal, we used two simple approaches. The first one
was based on the standard deviation (STD) in each channel, i.e., we assumed that the
lower the STD in the channel, the lower the noise. The second one used the amount
of white noise in each channel, estimated in the frequency range of 1 Hz to 40 Hz. In
both cases the estimations were performed using the EEG segments where stimulation
was not presented. To keep the results comparable to the baseline ones in terms of the
number of electrodes, we used the 3 electrodes with the lowest noise. The comparison of
the achieved accuracy is depicted in Figure 3. The figure clearly illustrates the benefits
 
Search WWH ::




Custom Search