Introduction

Steady-state visual evoked potentials (SSVEPs) are responses of the visual cortex to visual stimuli at specific frequencies. When the retina is excited by a visual stimulus, the brain generates electrical activity at the same frequency (and/or harmonics) of the visual stimulus. In the electroencephalogram (EEG), three bands can be identified: low- (up to 12 Hz), medium- (12–30 Hz), and high-frequency (≥ 30 Hz) (Zhu et al. 2010).

SSVEP response is often maximal in occipital area; consequently, this region is normally used to measure SSVEPs (Vialatte et al. 2010). In fact, generally, brain-computer interfaces (BCIs) based on SSVEP use electrodes located at occipital positions (O1, O2, and/or Oz) (Muller-Putz and Pfurtscheller 2008, Diez et al. 2011). However, this area is generally covered by hair, which causes bad impedance matching between the electrode contact and the skin (Wang et al. 2017). On the other hand, neuroscience studies based on EEG and other techniques (like MEG, PET, fMRI) reported that SSVEP response also occurs in other brain areas, including parietal, temporal, frontal, and prefrontal (Vialatte et al. 2010; Di Russo et al. 2007; Pastor et al. 2007; Floriano et al. 2018; Srinivasan et al. 2007, Sammer et al. 2005, Fawcett et al. 2004). However, after years of research, the complex mechanisms behind SSVEP distribution are not yet fully understood; details about proposed theories can be found in (Vialatte et al. 2010).

Thus, towards a more practical BCI, recent researches have analyzed SSVEP measured from below-the-hairline areas using some montages of EEG channels and frequency bands. For example, Norton et al. (2015) measured SSVEPs with an electrode positioned behind-the-ear with ear-referenced electrode. Hsu et al. (2016) obtained medium-frequency range SSVEP on the forehead with reference electrode on the behind-the-ear area. In a recent study using low- and medium-frequency band, it was found that behind-the-ear areas have better SNR in comparison to other hairless areas (Wang et al. 2017).

The reference electrode position and the stimulation frequency affect the SSVEP measured on the scalp (occipital area). However, the influence of these factors on SSVEP from below-the-hairline areas in the three frequency bands was not addressed. Thus, this study aims to investigate how reference electrode and the frequency bands (low, medium, and high bands) affect the SSVEP measured on hairless areas.

Materials and methods

EEG acquisition and experimental protocol

EEG signals were acquired with a Grass 15LT amplifier system and digitalized with a NI-DAQ-Pad6015. The sampling frequency was set at 256 Hz. The cutoff frequencies of the analog band-pass filter were set to 1 and 100 Hz. Additionally, a notch filter for 50 Hz (Argentinian power supply) was applied. Figure 1a presents the positions where the electrodes were located.

Fig. 1
figure 1

Left: positions on the scalp where the electrodes were located: (a) top view and (b) side view. Right: overview of the protocol: (c) the experiment protocol consisted of five runs; (d) each run was composed of 12 trials (one per each stimulation frequency)

Full size image

Twelve healthy subjects (ages 26.1 ± 4.1; 6 F and 6 M) with normal or corrected to normal vision participated in this study. The EEG recordings were conducted in a laboratory with low background noise and dim luminance. The study was approved by the Ethics Committee of the School of Exact, Physical and Natural Sciences of the National University of San Juan Argentina (act #7). Before participating in this study, all volunteers read an information sheet and signed a consent form.

Each subject sat in a chair at 60 cm from the stimulus. The experiment consisted of five runs, and each run was composed of 12 trials, one per stimulation frequency (Fig. 1c). The stimulation frequencies were presented in random order to each volunteer. Each trial lasted 7 s, with a variable separation between trials from 2 to 4 s, to avoid expectation effect (Fig. 1d). The trial begins with a beep (at t = 0 s) and 2 s later the stimulus is turned on. The stimulus stays on until the end of the trial at t = 7 s. At this moment, a feedback is presented to the volunteer, indicating whether the SSVEP was detected or not. The feedback is important to maintain the interest of the volunteer in the test. The subject was recommended relax for 2–5 min before beginning the next run. The advantage of using a short-time trial (7 s) in this study is due to potential real-time applications, such as, for example, control of a wheelchair.

The visual stimulus was composed of a light-emitting diode (LED) that illuminates a diffusion board of 4 cm × 4 cm. The LED can flick at different frequencies, from 5 to 65 Hz, with intervals of 5 Hz, comprising 12 frequencies, similarly as done by (Lin et al. 2012). Notice that 50 Hz was not used as stimulation frequency because this is the power line frequency in Argentina. Therefore, the stimuli range covers the three SSVEP bands (low-, medium-, and high-frequency).

In order to evaluate how the electrode montage affects the measurement of SSVEP, 19 channel configurations were evaluated, such as shown in Table 1. The ground electrode (GND) was placed at A2 (see Fig. 1b).

Table 1 Electrode montages
Full size table

The linked reference consists of a virtual reference obtained by averaging the potentials recorded at the left and right mastoids (LKT) or ears (LKE).

EEG data analysis

First, the EEG was digitally filtered with a Butterworth filter, order 6, bandwidth 3–70 Hz. Then, an EEG segment of 5 s was extracted between t = 2 s and t = 7 s and the magnitude of the frequency components of the signal was calculated based on the discrete Fourier transform (DFT) of the signal x[n], defined as

$$ F(f)=\left|{\sum}_{n=0}^{N-1}x\left[n\right]{e}^{-j2\pi fn{T}_s}\right| $$
(1)

where F(f) is the magnitude of the signal, Ts is the sampling period, N is the total number of samples of the signal, and f is the frequency.

The SNR was computed in (2) based on the values extracted from Eq. (1). The SNR of the SSVEP at a single channel is defined as the ratio of F(f) to the mean amplitude of the K neighboring frequencies (Wang et al. 2017):

$$ SNR(f)=\frac{K\times F(f)}{\sum_{n=1}^{K/2}\left[F\left(f+n\times \Delta f\right)+F\left(f-n\times \Delta f\right)\right]} $$
(2)

where ∆f is the frequency resolution (0.2 Hz in this study) and K was set to 8 (i.e., four frequencies on each side) (Chen et al. 2014). These parameters F(f) and SNR(f) were calculated using the 19 channels shown in Table 1.

Also, in order to have an overview of the distribution of the most relevant frequencies to each channel, the 12 frequencies (5–65 Hz) were ranked using the SNR value, according to (Müller-Putz et al. 2008, Müller et al. 2015). The frequency with the highest SNR received the score 12, the second the score 11, and so forth. Then, the scores of each frequency were summed up over the subjects (maximum is 144, i.e., 12 subjects × 12 (highest score)). The final result is the selection of a group of frequencies with the best SNR to each channel.

Statistical analysis

The one-way analysis of variance (ANOVA) was applied to the data. The statistical tests were run for each stimulation frequency, and then we analyzed the behavior of the SSVEP from every channel in each frequency. Following, the Tamhane T2 was used for post hoc tests.

Results

Figure 2 shows the mean amplitude of the SSVEP calculated at each stimulation frequency. The mean SNR of the three groups is depicted in Fig. 3.

Fig. 2
figure 2

Mean SSVEP amplitude of all volunteers at each stimuli frequency

Full size image
Fig. 3
figure 3

Mean SNR of the SSVEPs of all volunteers at each stimuli frequency. a Occipital group. b Temporal group. c Frontal group

Full size image

The best channel from each group is presented in Fig. 4, i.e., Oz-LKT (from the occipital group), Tp9-LKE and Tp10-LKE (from the temporal group), and Fpz-LKT (from the frontal group). Additionally, Fig. 5 shows the result of score evaluation to each channel.

Fig. 4
figure 4

The best channel for each channel group (temporal, frontal, and occipital)

Full size image
Fig. 5
figure 5

Sum of the scores of all volunteers for all channels

Full size image

Discussion

In our research, the SSVEP on Oz electrode was acquired for comparison purposes. Generally, the occipital group achieved higher SSVEP amplitudes and higher SNR than the other groups.

The temporal group presented the lowest amplitude values. Within the temporal group, between 5 and 20 Hz, Tp9-RE and Tp10-LE channels showed higher SSVEP amplitudes (Fig. 2). On the other hand, in evaluating the SSVEP amplitudes on the frontal group, no pattern was found, as all channels presented similar amplitudes at each frequency.

Figure 3b shows a peak at 35 Hz and, curiously, an increment of the SNR as the frequency increases. In this case, Tp9-LE achieved the best SNR for the range of 5 to 20 Hz, and Tp10-LKE presented the best SNR between 25 and 65 Hz.

In Fig. 4, Fpz-LKT presented higher SNR between 5 and 35 Hz than for the temporal group. On the other hand, from 40 Hz, the temporal group is a better option. Particularly, Tp10-LKE presented higher SNR values than Tp9-LKE.

The occipital group achieved higher SNR, particularly up to 40 Hz, which was statistically significant (p value < 0.05). On higher stimulation frequencies, this superiority is not so evident and, indeed, some non-hair positions achieved similar SNR.

Our findings are consistent to that found by Norton et al. (2015), which used low stimuli frequency (range of 6–10 Hz), as they reported high SNR for the channel closest to the reference on the ear. In our study, the behind-the-ear (Tp9-LE) position achieved a good SNR in low- and medium-frequency range (5–20 Hz) (see Fig. 3b). In addition, our results show a peak at 35 Hz and, curiously, an increment of the SNR as the frequency increases. On the other hand, Hsu et al. (2016) measured SSVEP on Fpz referenced to left mastoid (Tp9), with stimulation frequency from 13 to 31 Hz. They found similar SSVEP amplitudes to the ones depicted in Fig. 2. Moreover, they reported higher SNR between 17 and 21 Hz, decreasing further. Similarly, we obtained higher SNR between 10 and 20 Hz (Fig. 3c), but we found an additional peak at 35 Hz. However, different of our study, high-frequency stimulation (≥ 35 Hz) was not analyzed in Hsu et al. (2016), where the SSVEP presents a different behavior and the differences among channels are reduced. According to our results, a better result may be obtained if Fpz-LKT is used instead of Fpz-Tp9.

Conclusion

In this work, we presented a study about the characteristics of the amplitudes and SNR of SSVEP elicited on the three frequency ranges (low, medium, and high), which were evaluated on hairless areas. It was found that occipital area in fact presents the best SNR, but only up to 40 Hz. Beyond 40 Hz according our study, this superiority is not observed. On the other hand, the best electrode combinations for measuring SSVEP on hairless areas are obtained with Fpz-Tp9 and Fpz-Tp10 (up to 40 Hz). At frequencies higher than 40 Hz, the temporal positions referenced at the ear are the best options. As a contribution to the state-of-art, our study provides results that can aid in the setup of new SSVEP-based BCIs.