Dry andWater-Based EEG Electrodes in SSVEP-Based BCI Applications - Biomedical Engineering Systems and Technologies - page 11

(a) (b) (c)

1 1 1

0.9 0.9 0.9

0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6

0.5 0.5 0.5

0.4 0.4 0.4

Baseline
Lowest STD
Lowest White Noise
PSD Discrimination

0.3 0.3 0.3

0.2 0.2 0.2

12 15 18 21 28 32 36 40 12 15 18 21 28 32 36 40 12 15 18 21 28 32 36 40

Frequency [Hz] Frequency [Hz] Frequency [Hz]

Fig. 3. Classification accuracy for three setups with the optimal selection of 3 electrodes for PSD
computation: a) Dry contact electrodes; b) Water-based electrodes c) Conductive gel electrodes

of using the channels with the lowest noise for dry and gel electrodes, irrespective of
the approach used for noise estimation. The effect for water-based electrodes is not so
pronounced.

4.3 SSVEP Discrimination Power and Electrode Selection

The second feature that we explored for optimal electrode selection was the discrimi-
nation level of the PSD of the RVS during the stimuli period versus PSD of the RVS
during the non-stimuli period. For each stimuli period we selected the three electrodes
with the higher discrimination power for all 4 stimuli. The extent to which this approach
compares to the baseline and noise estimation approaches is also depicted in Figure 3.
Although it outperformed baseline results for dry and gel electrodes, the accuracy was
lower than for the noise estimation approaches across all three setups and even lower
than the baseline run for the water-based setup (see Figure 3b). Consequently, we in-
fer that for the optimal electrode selection it is of high importance to select the EEG
channels that have the lowest impact of noise in order to make a good estimation of the
SSVEP power. Thus, for the rest of the paper we use the optimal selection of electrodes
with the lowest white noise component.

4.4 Usage of Harmonics and Different Electrode Number

Figure 4a illustrates the effect of using first harmonics in the noise estimation as well
as PSD estimation on classification accuracy. For dry electrodes, significant increase
was observed only for the high frequencies, while the improvements of accuracy are
significant for both low and high frequencies of water-based and gel setups. With the
use of harmonics, the mean accuracy across subjects can be increased to more than
60% for dry setup (except for 12 Hz stimuli frequency), to more than 70% accuracy for
water-based electrodes, and to more than 88% accuracy for gel electrodes.
So far in the analysis we have used 3 electrodes as we reckoned that this number
would be sufficient to achieve good SSVEP detection performance. To test this hypoth-
esis and to investigate the impact of the number of electrodes used in the analysis, we