Detection of common EEG phenomena using individual electrodes placed outside the hair

Many studies over the past decades have provided exciting evidence that electrical signals recorded from the scalp (electroencephalogram, EEG) hold meaningful information about the brain’s function or dysfunction. This information is used routinely in research laboratories to test specific hypotheses and in clinical settings to aid in diagnoses (such as during polysomnography evaluations). Unfortunately, with very few exceptions, such meaningful information about brain function has not yet led to valuable solutions that can address the needs of many people outside such research laboratories or clinics. One of the major hurdles to practical application of EEG-based neurotechnologies is the current predominant requirement to use electrodes that are placed in the hair, which greatly reduces practicality and cosmesis. While several studies reported results using one specific combination of signal/reference electrode outside the hair in one specific context (such as a brain-computer interface experiment), it has been unclear what information about brain function can be acquired using different signal/referencing locations placed outside the hair. To address this issue, in this study, we set out to determine to what extent EEG phenomena related to auditory, visual, cognitive, motor, and sleep function can be detected from different combinations of individual signal/referencing electrodes that are placed outside the hair. The results of our study from 15 subjects suggest that only a few EEG electrodes placed in locations on the forehead or around the ear can provide substantial task-related information in 6 of 7 tasks. Thus, the results of our study provide encouraging evidence and guidance that should invigorate and facilitate the translation of laboratory experiments into practical, useful, and valuable EEG-based neurotechnology solutions.


Introduction
Over the past several decades, thousands of studies demonstrated that electrical signals recorded from the scalp (electroencephalogram, EEG) can provide meaningful information about a persons function or dysfunction.Some of this information is commonly used in clinical practice.For example, visual inspection of EEG recordings by experts provide useful diagnostic information to clinicians about specific sleep disorders (Rundo and Downey 2019) or about the nature and location of epileptic seizures (Sharbrough 1993).Other examples of potentially useful information in the EEG include parameters of depression (Neto and Rosa 2019), Alzheimer's disease (Jeong 2004), or attention (Hamadicharef et al 2009).Many of these studies take advantage of common EEG phenomena, such as evoked potentials in response to auditory or visual stimuli (e.g.Doan et al 2021).
Together, these studies provide overwhelming evidence that EEG holds information that, in principle, could prove useful not only in the context of research studies and diagnostic applications in a clinic, but also to a large number of people in their home.However, it is currently largely unclear how to transfer the potential benefits suggested or realized by these research or clinical studies to practical in-home solutions that improve the lives of many people.As a case in point, current EEG research and clinical studies often employ complex multi-channel montages that require laborious preparations by experts.Many such studies focused on the qualitative characterization of the complex spatio-temporal nature of a particular EEG phenomenon, such as the auditory evoked potential, during a specific task when the EEG is measured against a fixed reference (Goff et al 1977).Other studies attempted to quantitatively detect one specific EEG signal, such as in recent reports on EEG-based rehabilitation of motor or speech function (e.g.Musso et al 2022) or in the hundreds of reports on braincomputer interfacing experiments (Wolpaw et al 2002, Hwang et al 2013, or Abiri et al 2019, for review).Most of these studies employed complex EEG montages (presumably to maximize detection accuracy), although some focused on detection of one specific task-related signal detected at one particular combination of signal and reference location, usually in the hair (Ajami et al 2018, Angrisani et al 2018, Autthasan et al 2019).
In summary, the existing research and clinical EEG literature typically focuses on detection of EEG signals that are related to a specific task, that are detected using signal electrodes referenced against one fixed reference, and that are located predominantly in the hair.Thus, while informative in the context of a specific neuroscientific hypothesis, these existing studies have left a lot of room to learn more about the design of practical neurotechnologies that rely on detection of informative EEG phenomena in situations outside the laboratory.
The vision of the Chen Frontier Lab for Applied Neurotechnology is to bring neurotechnologies outside the laboratory and clinic and to make them useful to large populations.A key requirement to make such systems practical is to design them to rely on a minimal number of electrodes.These electrodes need to be easily applied and provide robust recordings, which can most readily be accomplished in easily accessible locations such as the forehead or in/around the ear and away from hair.It is important to note that it is almost certain that these constraints will reduce the type and amount of information that can be obtained.On the other hand, there is encouraging evidence that useful information can be obtained even with limited montages.While there have been studies that explored the utility of one or only a few electrodes placed outside the hair (e.g.Denk et al 2018, Ogino et al 2019), they usually only explored EEG responses in one task and for one specific placement of signal and reference electrode.
Thus, an important and currently largely unanswered question is the degree to which common EEG phenomena related to auditory, visual, cognitive, motor, and sleep function can be detected from different combinations of individual signal/referencing electrodes that are placed outside the hair.
To answer this question, in the present study, we asked 15 subjects to engage in seven tasks while we recorded EEG signals from two locations on the forehead and three locations around the ear.For comparison, we also recorded from four additional locations in the hair.We then used quantitative metrics to illustrate the extent to which task-related EEG phenomena can be detected by the different electrodes when they were re-referenced to all possible reference electrodes.The results show that signals from locations on the forehead and around the ear were surprisingly informative about most tasks.

Subjects and data collection
In this study, 15 healthy subjects (9 females, 6 males, age: 29.3 ± 4.2 years, normal or corrected-to-normal vision) participated in 7 different tasks.All subjects provided written consent for participation in the study, which was approved by the ethics review board of Huashan Hospital.
We recorded EEG signals using the general-purpose software BCI2000 (Schalk et al 2004, Schalk andMellinger 2010) that we interfaced with a 128-channel bio-signal amplifier (g.HIamp, g.tec, Graz, Austria), and 2 motion tracking sensors (MTw Awinda, Xsens, The Netherlands) that were placed on the subject's arm and the leg, respectively.The bio-signal amplifier acquired EEG signals from 9 scalp locations that were referenced to the right earlobe (location A2 according to the extended 10-20 system described in Sharbrough et al 1991).See figure 1 for details.
Our study focused on locations outside the hair that are easily accessible.Only two general areas (the forehead and locations around the ear) meet these requirements.We selected five locations from these two areas.Four of the five electrodes (Fp1 and Fp2 on the forehead and A1 and M1 around the ear) are commonly used in EEG montages.Based on evidence that electrodes around the ear can detect differential information (Denk et al 2018) we also added another location around the ear (labeled here as E1 and shown in figure 1).While not the primary target of our study, we supplemented these 5 locations with four electrodes placed in other common locations in the hair (F4, C4, Fz, and Cz) to provide additional context.
Throughout the experiment, we strove to keep electrode impedance below 10 kΩ and sampled the resulting EEG and motion tracking sensors at 256 Hz.

Behavioral tasks
All subjects performed 7 tasks (T1-T7, figure 2) in a dimmed room, and (all except for T7) in front of a 27" LCD screen at a distance of approximately 60 cm.The duration of the tasks varied between 2-60 mins (T1: 2 min, T2: 7 min, T3: 3 min, T4: 3 min, T5: 3 min, T6: 3 min, T7: 60 min).There was a break of 1 minute (T1-T5) or 15 minutes (T6) after each task.The total experimental duration was approximately 120 mins including the time for electrode placement.The tasks are summarized in figure 2, and are described in more detail below.
In Task 1 (T1), eyes open/closed, the subjects were first asked to rest with their eyes open (1 min) and then to rest with their eyes closed (1 min).
In Task 2 (T2), mental calculation, the subjects rested for 2 seconds while the screen showed the word 'REST.'Then, the subjects prepared for the task for 3 seconds while the screen showed a '+' sign.Finally, the subjects were presented with a mathematical equation (addition of two 3-digit numbers, e.g.899 + 917 = 1816).The subjects were asked to judge whether the equation was correct or not by indicating their response with a button press.The system proceeded to the next trial after the button press or after a time-out of 10 seconds, whichever came first.There were a total of 50 trials.
In Task 3 (T3), hands opening/closing, the subject was asked to repeatedly open and close both hands while the screen showed the words 'open/close hands' for 5 seconds, and then to keep both hands opened and relaxed while the screen showed the word 'relax' for 5 seconds.This sequence repeated 18 times and hence lasted a total of 3 minutes (5 s * 2 * 18 = 180 s).
In Task 4 (T4), sound stimulation, the subjects were asked to listen to two types of sounds in an oddball paradigm.The two types of sound consisted of a sine wave at 1024 Hz (standard stimulus), and a sine wave at 2048 Hz (deviant stimulus), respectively.Each sound stimulus lasted for 100 ms.The inter-stimulus interval (ISI) randomly varied between 350-450 ms.In total, there were 360 sound stimuli (288 standard, 72 deviant).The sequence of stimuli was block-randomized in blocks of 10.The subjects were asked to count total number of deviant stimuli presented throughout the task.
In Task 5 (T5), visual stimulation with images, the subjects were presented with blocks of images; each block had 50 images.The images consisted of images of plants (standard stimuli) and faces (deviant stimuli).The sequence of the images was block-randomized, and each image was presented for 300 ms.The interstimulus interval (ISI) randomly varied between 150-250 ms.In total, there were 350 images (280 standard, 70 deviant).The subjects were asked to count total number of face images presented throughout the task.
In Task 6 (T6), visual stimulation with LEDs, the subjects were asked to close their eyes and rest while two LEDs simultaneously flashed on (100 ms) and off (300-500 ms) in front of their eyes continually for 3 mins.The LEDs were mounted on the computer screen (with each LED approximately in front of each eye).
In Task 7 (T7), the nap session, subjects were given a 60-min afternoon nap opportunity.They were asked to lie on a reclining chair or a folding bed, and to attempt to fall asleep in a dim and air-conditioned room.Immediately after the nap, they were asked to self-report on their sleep including questions about whether they had fallen asleep and the rough estimates on sleep latency and sleep duration.Thirteen of the subjects reported that they fell asleep during the experiment.The mastoid electrode dropped off about 10 mins into the nap for another subject, so we included 12 subjects in our analysis of T7.

Signal pre-processing
Our signal pre-processing began with generating all 36 possible combinations of re-referenced signals by simply subtracting the signal from one electrode from the signal from another electrode.For tasks T1-T6, we then subjected each of the 36 signals to bandpass filtering (0.3-40 Hz, 3rd order Butterworth filter) to eliminate low-frequency drifts and line noise.
Our analysis for T7 focused on delta (1-4 Hz band as suggested in Purves et al 2001) activity, one of the most characteristic signals associated with slow-wave sleep.However, delta activity can be contaminated by movement or eye blink artifacts.To minimize the potential effects of these artifacts on our analysis, we excluded periods of movements and eye blinks.We detected movements by first bandpass filtering the three dimensional rotational signals (roll, pitch, and yaw) that we acquired from each of the two movement sensors (6 movement signals in total) with a 2nd order highpass Butterworth filter and a 4th order lowpass Butterworth filter (cutoff frequency at 1 Hz and 10 Hz, respectively).We then calculated the root-meansquare of the 6 filtered movement signals and excluded the EEG data in the periods when the root-meansquare value of the 6 movement signals exceeded a manually-defined threshold of 2.4 degrees.We then bandpass-filtered the remaining EEG signals with a 2nd order highpass Butterworth filter and a 4th order lowpass Butterworth filter (cutoff frequency at 1 Hz and 35 Hz, respectively) and epoched the EEG data into 4 s periods with 50% overlap.We rejected all periods if signals at Fp1 or Fp2 had amplitudes greater than their respective mean+5×standard deviation or if any other EEG channel had signals with amplitudes greater than their respective mean+8×standard deviation.This eye-blinking rejection procedure rejected about 6% of the epochs.Visual inspection confirmed that the remaining data were practically free from artifacts that could affect our analyses.
For all analyses that produced the results presented in figures 3 and 4, we then epoched the bandpass-filtered re-referenced signals as follows: • T1:We cut the data from the eyes-open or eyesclosed conditions into 2 s epochs, and labeled them Task or Rest epochs, respectively.
• T2: We cut the data into Rest epochs (from 0.5 s after the '+' appeared to 0.5 s before the '+' sign disappeared) and Task epochs (from the time the equation appeared to 0.5 s before the equation disappeared).
• T3: We cut the data into Rest epochs (4 s after the 'relax' text appeared) and Task epochs (the 4 seconds after the 'open/close hands' text appeared).
• T4: We cut the data into Standard epochs (0.1 s before to 0.3 s after the onset of the standard stimulus) and Deviant epochs (0.1 s before to 0.3 s after the onset of the deviant stimulus).We eliminated all epochs in which signal amplitude at any time and for any electrode/reference combination exceeded 100 μV.
• T5: We cut the data into Standard epochs (0.1 s before to 0.3 s after the onset of the standard stimulus) and Deviant epochs (0.1 s before to 0.3 s after the onset of the deviant stimulus).We eliminated all epochs in which signal amplitude at any time and for any electrode/reference combination exceeded 100 μV.
• T6: We cut the data into epochs from 0.1 s before to 0.3 s after the onset of the LED flash.We eliminated all epochs in which signal amplitude at any time and for any electrode/reference combination exceeded 100 μV.
• T7: We cut the data into 4-s epochs with 50% overlap and rejected all epochs with eye-blinking artifacts detected based on our rejection criteria as described above.

Statistical analysis
We then set out to illustrate examples of EEG responses to the seven tasks in individual subjects.In these examples, which are shown in figure 3, we calculated average time courses and spectra across all epochs for T1-T6, and average power of delta activity for T7.To highlight the statistical difference between the time courses or spectra corresponding to two conditions, we calculated, for each point in time or each frequency, the coefficient of determination r 2 (Winer 1962) between the two distributions of voltage or frequency amplitudes corresponding to the two conditions.
To produce the main results of our study, we used quantitative metrics to describe the degree to which each re-referenced signal held information about the task.
For the experiments that resulted in different frequency responses (T1, T2, and T3), our metric was classification accuracy, which we calculated for the Task and Rest epoch.Specifically, we first calculated EEG spectra (T1: 4-20 Hz, T2: 4-20 Hz, T3: 0.3-40 Hz), separately for task and rest conditions, concatenated them across all epochs and the two conditions, labeled them with 0 and 1 according to task and rest, respectively, and finally determined the accuracy of a linear discriminant function that attempted to separate the data for the two different conditions using Matlab's fitcdiscr function (10fold cross-validation, accuracy due to chance was 50%).
For the experiments that induced time-locked amplitude responses (T4, T5, and T6), our metric was the signal-to-noise (SNR) measurement described in Schalk et al 2007.The SNR metric determined the degree to which auditory/visual stimuli produced consistent EEG responses.We determined this consistency by submitting the EEG time courses to the SNR procedure after downsampling the EEG signal by a factor of 4 (i.e. to 64 Hz).This procedure is highly sensitive to amplitude modulations following the stimulus and relates the inter-sample variance across the response detection period, σ 2 ( f ), to the average within-sample variance, , resulting in a single SNR value for each re-referenced electrode.SNR values close to 1 mean that the stimulus produces no consistent modulation in the EEG signals.
For the nap experiment (T7), we were primarily interested in the degree to which delta (1-4 Hz) activity, which is well known to increase in amplitude between the resting condition and deep sleep (Purves et al 2001), changed throughout the experiment.Thus, our quantitative measurement was calculated simply as the variance in delta activity, measured across all signal samples in the nap.Specifically, we first calculated the power in the delta band for each 4 s EEG epoch using Welch's method.We then averaged these power estimates across one minute, calculated the variance of these averages across all available minutes, and normalized those values within each subject to account for individual differences in delta variance.Finally, to highlight the degree to which the delta signal at F4-M1 (i.e. the EEG derivation recommended for detection of sleep-related delta activity by the American Academy of Sleep Medicine (AASM) (Berry et al 2018)) could also be detected in other signal derivations, we calculated the squared correlation coefficient (r 2 ) between the signal at F4-M1 and the signal at other derivations.
To produce the main results for our study, we averaged accuracy, SNR, delta variance or r 2 values across subjects, and did so for each of the 36 re-referenced signals.

Results
The main results of this study are shown in figure 4 for each of the seven tasks.
For T1, eyes open/closed, our results demonstrate that the brain exhibits changes between these two conditions that can be detected well (accuracy between 62% and 76%) in all re-referenced signals-all these accuracies are statistically better than chance (p < 0.05, bootstrap randomization test).These results are consistent with the finding that different physiological processes in the brain change between the states of eyes open and eyes closed, and can be detected at different location and different frequencies (Barry et al 2007).Visual inspection of the spectral differences between the two tasks (such as in figure 3-T1) confirmed that our results were not driven by artifacts such as eye blinks.Brain signals related to this task can readily be detected in many combinations of frontal locations and/or locations around the ear.
For T2, mental calculation, our results show that the brain produces signals that can differentiate mental calculation from rest.These signals can best be detected (p < 0.001, t-test) by frontal locations (i.e.average accuracy = 66% for FP1 and FP2 referenced to any other location) compared to other locations (i.e.average accuracy = 59%).These results are consistent with previous findings that described such signals (like frontal-midline theta) in prefrontal locations (Sasaki et al 1996).Finally, signals related to this task can readily be detected in frontal locations referenced to locations around the ear.
For T3, hands opening/closing and resting, we found that some re-referenced signals measured at C4 and Cz (i.e.locations close to hand areas of motor cortex) detected the differences between hand movements and rest well (average accuracy of 69%).This finding is expected and consistent with the numerous descriptions of the typical topographical distributions of movement-related mu/beta rhythms (McFarland et al 2000).At the same time, our results also demonstrate that these movement-related signals cannot be detected at all (p > 0.05, bootstrap randomization test) or only marginally (average accuracy of 54%) by any locations outside the hair.
For T4, auditory stimulation, our result show that, not surprisingly, signals related to deviant stimulation can be more easily detected than signals related to standard stimulation (i.e.comparing SNR values for the top-left triangle in figure 4) with those in the bottom-right triangle: p ≪ 0.001, paired t-test).They also show that brain signals related to this task can readily be detected in many combinations of frontal locations and/or locations around the ear (SNR values >1.015 are significant at the 0.05 level as determined by a bootstrap randomization test).
For T5, visual stimulation with images, our results are similar to those for auditory stimulation (i.e.comparing SNR values for the top-left triangle in figure 4) with those in the bottom-right triangle: p ≪ 0.001, paired t-test).They also confirm that brain signals related to this task can readily be detected in many combinations of frontal locations and/or locations around the ear.
For T6, visual stimulation with LEDs, our result suggest that visual stimulation with LEDs leads to evoked potentials even when the eyes are closed (consistent with Pojda-Wilczek et al 2019), and that those evoked potentials can be detected in frontal locations (FP1 and FP2) better than standard responses during T5, visual stimulation with eyes open (p = 0.003, paired t-test).Moreover, the responses at FP1 referenced to FP2 can be detected better than at all but four other combinations of signal/reference locations.
For T7, delta activity during the nap experiment, our results are in line with existing literature that delta activity (as indexed by signal variance) is mostly predominant over fronto-central derivations, such as F4-M1 or C4-M1 as recommended by AASM (Berry et al 2018).Our results also provide encouraging new quantitative evidence that such information can also be detected using electrodes outside the hair (e.g.Fp1-M1 or E1-M1).Specifically, while the variance of delta activity across the nap experiment at these derivations is rather small (results in top-left half), they are closely correlated with the signals (of much larger amplitude) detected at F4-M1 (r 2 results in bottom-right half).

Findings in this study
In our study, we set out to determine to what extent EEG phenomena related to seven different tasks can be detected in different combinations of signal/reference electrodes, focusing specifically on locations on the forehead and around the ear that are not covered by hair.The data shown in figure 4 support five main conclusions that are summarized below.
First, EEG signals at different combinations of 'hair-free' signal/referencing locations can readily detect EEG phenomena related to six of the seven tasks.At least in its totality, this result is somewhat unexpected, as the bulk of the literature studied these phenomena using traditional full-head 16-64 channel montages (often referenced to the ear), and typically discussed the location of the largest magnitude of the signal (often in lateral, central, or occipital locations).The finding in our paper suggests that only very few electrodes placed on the forehead and/or around the ear could enable detection of different types of brain signals, and thus may support different neurotechnology applications.Thus, the finding described in our paper increases the probability that these task-related signals could provide the basis for neurotechnologies that can be rapidly applied (perhaps using rapid-application electrodes similar to those used in ECG applications), and hence may be more practical than those relying on the more traditional and cumbersome types of electrodes placed in 'hair' locations.
Second, when considering 'hair-free' electrodes on the forehead, locations around the ear may be necessary to optimize detection performance.
Third, signal derivation Fp1-Fp2 showed only little information.This appears to include the detection of delta activity in the nap task.On the one hand, this finding suggests that the increasing number of devices that promise to provide sleep-related information using locations only on the forehead may not detect all important components of sleep-related signals.At the same time, there is still a substantial relationship between delta recorded at this combination of locations with the more traditionally used F4-M1 montage.
Fourth, different derivations of A1, M1, and E1 were surprisingly informative about different tasks (T1, T4, T5, and T6).This suggests the possibility that locations in/around the ear may provide sufficient information to support certain applications.
Fifth, detection of signals related to motor movements cannot meaningfully be accomplished from locations outside motor cortex.On the one hand, this is unsurprising and consistent with the hundreds of previous studies that showed that mu/beta rhythms can best be detected around sensorimotor cortex.On the other hand, since these signals practically cannot be detected in 'hair-free' locations, brain-computer interface (BCI) systems that rely on the mu rhythm will most certainly have to continue to rely on electrodes placed in central locations and thus, in the hair.This requirement appears to substantially reduce the possibility that BCIs based on motor imagery will be able to support applications other than those that provide substantial benefits (such as neurorehabilitation in people with chronic stroke (Bundy et al 2017)).

Implications
These results have important implications for the design of EEG-based neurotechnologies that may be useful for people outside typical laboratory or clinical environments.Despite important advances over the past decade, such as the increasing availability of convenient consumer EEG headsets (devices from Emotiv, g.tec, Muse, Neurosky, or OpenBCI), dry electrodes (Popescu et al 2007, Oehler et  Our study extends these reports by providing a more comprehensive evaluation of different locations, tasks, and referencing methods.Its results now more forcefully argue for the possibility of designing EEG-based neurotechnologies that can be applied rapidly, can function more reliably, may provide value for different applications, and may even be useful in situations outside the laboratory and clinic.

Potential shortcomings
While our results are encouraging, we focused primarily on the ability of common EEG phenomena rather than on their use in specific use cases (such as sleep monitoring, EEG-based rehabilitation, etc.).Thus, future research could provide more emphasis on the application context.For example, the results presented in the present study for the nap study (T7) are in line with existing literature that delta activity is mostly predominant over the fronto-central derivations such as F4-M1 or C4-M1, but also provide encouraging new quantitative evidence that such information can also be detected using electrodes outside the hair (e.g.Fp1-M1 or E1-M1).However, given the information in our study, we cannot definitively conclude that this information is also equally useful for a specific purpose such as sleep staging.

Summary
In summary, in our study, we asked 15 subjects to engage in 7 different tasks while we recorded EEG signals from 5 locations outside the hair and 4 auxiliary locations within the hair.Quantitative metrics of detection performance suggest that signals detected in locations outside the hair can detect task-related EEG signal changes in 6 of 7 tasks.

Conclusions and outlook
Neurotechnologies have the potential to improve many people's lives, but to date, we have barely scratched the surface of these opportunities.An important reason for this lack of widely accessible solutions is the current predominant requirement to detect EEG responses with complex multi-channel montages that include electrodes in the hair.The findings described in this paper suggest that it may be possible to develop neurotechnologies that can detect useful EEG signatures in different tasks using only two locations (one signal and one reference) outside the hair.Thus, future work may explore the detection of these signatures in specific contexts (such as sleep or evaluation of cognitive performance), which may eventually lead to useful and practical neurotechnologies for everyone.

Figure 1 .
Figure 1.Recording locations.We recorded EEG signals from nine locations that were referenced to the right earlobe (A2).

Figure 2 .
Figure 2. The seven tasks and their sequence within the experiment.

Figure 3 .
Figure 3. Example r 2 time courses/spectra and voltage time courses/spectra for tasks T1-T6, time courses of delta power for task T7, given for all combinations of signal and referencing locations.T1: Results for eyes open/close for Subject 9. T2: Results for mental calculation for Subject 15.T3: Results for hands opening/closing and resting for Subject 15.T4: Results for auditory stimulation for Subject 15.T5: Results for visual stimulation for Subject 15.T6: Results for LED stimulation for Subject 15.T7: Results for the nap experiment for Subject 1.The unit bars and arrows in the bottom-left/upper-right corners indicate the analysis method and unit for the results presented in the upper-left or lower-right half, respectively.

Figure 4 .
Figure 4. Detection results for all tasks and for all combinations of signal and referencing locations.T1: Results for eyes open/closed.T2: Results for mental calculation.T3: Results for hands opening/closing and resting.T4: Results for auditory stimulation.T5: Results for visual stimulation.T6: Results for LED stimulation.T7: Results for the nap.Results in the bottom-right half give r 2 values for each signal derivation versus F4-M1.The unit bars and arrows in the bottom-left/upper-right corners indicate the analysis method and unit for the results presented in the upper-left or lower-right half.Accuracy due to chance was 50%.All SNR values >1.015 were significant at the 0.05 level as determined by a randomization test.
al 2008, Grozea et al 2011, Zander et al 2011, Xing et al 2018, Li et al 2021; or devices from Wearable Sensing), active electrodes (Laszlo et al 2014, Xu et al 2017; or devices from Biosemi or g.tec), the common requirement for placing electrodes in the hair makes preparation more time consuming, signal quality more unstable, and decreases cosmesis and/or practicality.While there have been several encouraging reports that focused on locations outside the hair (Mikkelsen et al 2015, Bleichner et al 2016, Papalambros et al 2019), these reports were typically focused on one specific location (such as the ear), one task, or one referencing method.