Dry andWater-Based EEG Electrodes in SSVEP-Based BCI Applications - Biomedical Engineering Systems and Technologies - page 8

For the gel electrode evaluation segment, the experimenter positioned the EEG cap
on a participants head and filled the holes (electrodes) in the cap with conductive gel.
The experimenter controlled the EEG signal quality and if needed added more gel to
improve the contact and achieve desired EEG signal quality level. This activity was
also timed. Then the EEG signal was recorded in the two runs, the same way as for
water-based and dry electrodes. After the recording, the EEG cap was removed from
the participants head. On completion of this session, a hair wash coupon was given to
the participant.
In the debriefing segment (only at the end of the second session) participants had
to fill in a questionnaire about their experience with the different electrode types and
mounting systems. They were also encouraged to give general comments on the setups
and the study design.
Signal Analysis. In the SSVEP BCI framework, the goal of signal processing methods
is to detect the presence of an SSVEP at a given stimulation frequency in the EEG. In
general, the problem consists of deciding if within a certain time window, the attention
of the subject on an RVS has been sufficient to elicit an SSVEP response. The main
challenge is to avoid the impact of various artifacts and noise (including background
EEG) on the SSVEP, as well as the selection of the best components (e.g., electrodes,
temporal segments) that contribute the most in the SSVEP response. These two aspect
correspond to artifact handling methods and algorithms for optimal SSVEP detection.
To minimize the impact of severe artifacts, expected when using dry and water-based
electrodes, we employed an algorithm for rejecting epochs with artifacts. We selected
the epoch duration of 1 s with 75% overlap. The algorithm excluded the epochs where
the absolute amplitude peak inside the epoch was larger than the empirically selected
threshold. The thresholds were estimated based on the standard deviation within the
recorded segment. We used a numeric value that is 5 times larger than the standard
deviation of the signal in each electrode. Such threshold was used for all three electrode
types. In addition, the standard deviation of the recording was used in estimating the
level of noise in a particular channel (see Section 4).
The strength of SSVEP can be estimated using various methods, ranging from uni-
variate based power spectral density (PSD) estimation to the use of multivariate spatial
filtering [9]. Since our intention was to compare the SSVEP strength measured with
different electrode setups and to infer the difference in performance, we employed a
PSD estimation method using the Welch algorithm [42], which also provided us with
easily traceable and interpretable results. The PSD was estimated on one-second long
epochs with 75% overlap (that do not contain artifacts). For selecting the best channels,
the absolute PSD value across the 1 to 40 Hz frequency spectra was used to estimate the
presence of noise in the EEG (see Section 4).
Evaluation Protocol. BCI performance is usually assessed in terms of classifica-
tion accuracy, classification speed, and the number of available choices. In SSVEP-
based BCIs, the classification accuracy is primarily influenced by the strength of the
SSVEP response, the signal-to-noise ratio (SNR), and the differences in the properties
of the stimuli. That is why we focus on reporting the accuracy of three setups. As the