Dry andWater-Based EEG Electrodes in SSVEP-Based BCI Applications - Biomedical Engineering Systems and Technologies - page 9

(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

0.3 0.3 0.3 Pp 1, short thin sparse hair

Pp 2, short thin sparse hair
Pp 3, medium thin dense hair
Pp 4, medium thick dense hair
Pp 5, long thick sparse hair
Pp 6, long thick dense hair

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0

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. 2. Classification accuracy for three different setups with 6 participants (Pp 1 to Pp 6) when
using PSD at the occipital sites (i.e., O1, O2, and Oz): a) Dry contact electrodes; b) Water-based
electrodes c) Conductive gel electrodes

classification speed depends on the time it takes for the SSVEP to be of sufficient
strength, we also report the information transfer rate (ITR). ITR is estimated using the
approach detailed in [43].
In addition to the bit rate, it is also important to consider the safety and comfort of
SSVEP-based BCIs. That is why participants had to rate the comfort level of each of the
setups, and provide additional information on whether and under what circumstances
would they use a particular setup.

4 Signal Quality and Performance

This section discusses the performance of dry, water-based, and gel-based setups con-
sidering the presence of noise in the EEG signal, usage of harmonics, and the optimal
selection of electrodes.

4.1 Baseline Performance and the Presence of Noise

Accuracy When Using Occipital Electrodes. We expected that the SSVEP response is
strongest in the occipital sites and therefore as a baseline measurement we selected three
electrodes at occipital sites, namely, O1, O2, and Oz. Figure 2 depicts the classification
accuracy (on 5 s-long segments) when maximum PSD in these electrodes is used for
each of the participants and for each of the setups. The accuracy for dry electrodes
(Figure 2a) is rather low, ranging from the chance level, i.e., 25% for participants with
long and/or thick hair (Participant 4, 5, and 6) to more than 50% for participants with
shorter and/or tinner hair (Participant 1, 2, and 3). These baseline results demonstrate at
the first stage of our analysis the problem of measuring the SSVEP response, and EEG
in general, using dry electrodes with people with long and thick hair.
The detection of 12 Hz response is challenging for water-based and gel electrodes, as
illustrated in Figure 2b and Figure 2c. This demonstrates the problem of distinguishing
the changes in the alpha power domain due to the overlap with the dominant alpha