unobtrusive ambulatory eeg using a smartphone and flexible printed electrodes around the ear - touch screen head unit

This study provides the first evidence that reliable EEG data can be recorded with a new cEEGrid electrode array printed on flexible paper and arranged in c-
Suitable for the shape around the ear.
Ten participants wore two cEEGrid systems for at least 7 hours.
Use a smartphone for stimulus delivery and signal collection, and collect resting EEG and auditory oddball data 6 to 7 hours in the morning and afternoon.
Analysis of confirmed static EEG data
Known spectral differences under open and closed eyes conditions.
ERP results confirm significantly greater predictive conditional effects of the target's ERP amplitude compared to standard tones, as well as high tests
Retest reliability of P30 amplitude (ru2009>u2009=u2009. 74).
In addition, the linear classifier trained according to the data of the morning meeting shows that the classification accuracy of the morning and afternoon meetings is similar (bothu2009>u200970%).
These findings demonstrate the feasibility of obtaining hidden and comfortable brain activity over multiple hours.
Twelve healthy volunteers who did not have past or present neurological or mental conditions participated in the experiment.
The sample consists of undergraduate or graduate students or staff from the ordenburg department of psychology, most of whom have previous experience with EEG.
There are two people in the analysis.
One is excluded due to technical issues, resulting in loss of EEG signals, and the other data set must be excluded because participants do not count pitch as per task instructiondeviant tones.
The final sample consists of 10 participants (
Aged 23-47; mean 29. 9 years; 5 male).
All participants received informed consent.
The Oldenburg University Ethics Committee approved the research programme, and all procedures were carried out in accordance with the approved programme.
In order to record still EEG data, participants were instructed to relax and keep their eyes open or closed, according to verbal instructions given by the experimenter.
For the oddball task, two pure tones (600, 900u2009Hz)
Duration 62 kbps ms (
Including the rise and fall time of 10 kbps ms)
With consumersHeadphones (
Samsung EHS64AVFWEG)
In the participants-
Control, comfortable loudness.
The standard tone is 600 hz and the target tone is 900 hz.
Set the target probability to 20% and give the target and standard in random order, the constraint target cannot be tracked directly to each other. A fixed inter-
Using a stimulus interval of 1000 kbps ms, an average of 860 stimuli were presented per session.
The participant's task is to silently count the high target tone while keeping his eyes wide open.
In order to ensure the participant's continued attention, the stimulus presentation was randomly interrupted two to four times during the recording session, and the participant had to report the number of target tones for the silent count.
Demo with the back of the robot
End of OpenSesame running on Sony Z1 Z1 smartphone (model: C6903; OS: Android 4. 4. 4).
Android operating system supports real-time demands.
The initial timing test showed that the audio presentation was not fully synchronized with the event markers sent to the EEG acquisition application.
This leads to an ongoing delay, which is moderately unpredictable in the recording session, but can be compensated when offline (see below).
Although rare time anomalies occur in less than 3% of all stimulus presentations, in the pilot test, when event markers are written after the sound presentation, the time between the marker time and the stimulus presentation
Delay jitter about 6 ms standard deviation).
With this procedure, all marks are given a constant value (100 samples).
Participants arrived in the morning and two cEEGrid devices were installed ().
This includes the preparation of skin around the left ear and right ear using grinding gel and alcohol swab. Double-
A tape is attached to CEEGrids, a drop of electrolyte gel (
Abralyt HiCl, Germany Easycap Limited)
Is added to the conductive surface and then cEEGrids is attached to the skin around the ear.
The average preparation is less than five minutes.
After the electrode is installed, connect the two cEEGrids to an improved miniature amplifier (
See below for more information)
, It is placed on the headband at the back of the head.
Impedance values in the range of 10 to 30 k Ω are available at a resolution of 5 k Ω.
When the impedance value is> 30 k Ω, each subject appears on average at less than 2 electrodes, adding another drop of gel using a blunt needle without removing cEEGrid.
Subsequent resting EEG data (
2 u2009 × u2009 1 u2009 min open your eyes 2 u2009 × u2009 1 u2009 min Close your eyes alternating arrangement)
And typical EEG (
Eye flashing, lateral eye movement)
The experimenter sitting next to the participant was recorded while holding the smartphone.
For still EEG records, the touch screen pressing performed by the experimenter adds an event marker indicating different situations (
Eyes wide open, eyes closed, artificial products).
Record the electrode impedance after the completion of the resting EEG task.
Participants then performed an auditory oddball task lasting about 20 minutes.
The amplifier was subsequently removed, two cEEGrid connectors, one from the left ear and one from the right ear, tied together to prevent them from swaying from the head.
In the afternoon, participants returned to the office and cEEGrids reconnected to the amplifier.
The same recording scheme was repeated, but in reverse order, that is, oddball data was recorded first, and then still EEG data was recorded.
The impedance was then checked.
Please note that from the first recording in the morning to the end of the second recording later in the afternoon, participants followed their normal daily activities, which included office and laboratory work, in most cases, it includes lunch break, walking and chatting with friends or colleagues.
During this period, participants were not monitored and received no clear instructions except to avoid pulling cEEGrids when wearing glasses or headgear.
Between morning and afternoon meetings, no EEG was recorded and the experimenter did not manipulate cEEGrid in any way.
Average 6 hours and 14 minutes (
6: 00, up to 7: 21)
Pass between completing the first odd task in the morning and starting the second one in the afternoon.
Therefore, the average wear time of cEEGrids is between 7 and 8: 20.
All recordings are made in a relatively quiet office environment.
After the afternoon meeting, participants are encouraged to provide feedback on wearing cEEGrids and specifically encourage them to report potential discomfort.
Finally, cEEGrids were removed and participants received an tissue to remove the residual conductive gel. A SMARTING 24-
A Channel Mobile EEG amplifier is used (
(Bergel, Serbia).
The system has a sample rate of 500 hz, a resolution of 24 bits and a bandwidth from DC to 250 hz (SMARTING, www. mbraintrain. com).
The amplifier unit includes a 3D gyro and power supply for several hours (64 grams of weight;
Size 82 u2009 × u2009 51 u2009 × u2009 14 u2009 mm)
And transmit data wirelessly via Bluetooth (v2. 1)
Protocol for matching devices located nearby.
The SMARTING amplifier comes with two safety figures (SD)
A memory card slot connected to two cEEGrids.
CEEGrids is designed in half
One-time equipment for the first author (S. D. )
Winter medical systems(
TMSI of Oldenzaal, Netherlands).
Flexprint materials include several layers of biocompatible polyamine.
The conductive part includes gold plated end, pure copper trace and conductive Ag/AgCl based polymer film ink.
The conductive surface is round, with a diameter of 3mm, and the distance between the electrodes located within cEEGrid is 12 or 18mm (
From center to center).
Number of poles (10)
And the size and shape of the cEEGrid version 1.
The 0 used in this study was recorded by the pilot and previously more around the ear
Channel EEG recording was performed using a micro-sintering Ag/AgCl electrode.
Signal acquisition is carried out in an application running on Android, provided by the amplifier manufacturer (
Smarting version 1. 0).
Data from 18 EEG channels and event markers are written to a common file format (. bdf).
Markers are generated by touch screen pressing for still EEG recording and opensession used to indicate sound presentation in the oddball paradigm.
Analysis of EEG data
Use the lines of EEGLAB version 13. 4.
4b and BCILAB version 1.
Custom scripts running under 1 and Matlab 7 (
Mathworks, Natick, MA).
The two electrodes in the middle of the right ear are used as grounding and reference.
Left Ear channel without homologous (i. e.
Ground electrode for right ear)
Discarded, the rest of the Channel
Refer to the mastoids linked by algebra, resulting in symmetric 16-
Channel montage (
8 channels per ear).
High EEG analysis data for rest
Filter with zero phase finite pulse response filter at 1 hz (−6u2009dB cut-off at 0.
5 hz, filter order 1650)
For each condition, segments with a length of 1024 samples and an overlap of 256 samples are Hanning windows submitted to the fast Fourier transform (
Pwelch method implemented by Matlab
, Mean and log-normalized (10*log10).
Visual inspection found that single channel (L1)
Due to unstable electrodes, the afternoon EEG of a single subject did not produce a reliable signalskin impedance.
The corrupted spectrum on this channel is replaced by the corresponding values in the same channel recorded in the morning session.
For the analysis of oddball data, there is no similar problem.
Further analysis of the results of the static EEG ensures that poor channel replacement does not affect the results of statistical evaluation.
For the analysis of oddball data, continuous data is a finite impulse response filtered out from 0. 2u2009Hz (−6u2009dB cut-off at 0.
1 hz, filter order 8250)to 20u2009Hz (
-6 db cutoff at 22. 5 hz, order 331)
The period from-200 to 800 ms is then extracted and baseline correction is performed (−200 to 0u2009ms).
Use the probability and peak criteria implemented in EEGLAB to identify an era dominated by artifacts (
Standard deviation: 2)
Further analysis was rejected.
In order to compensate for the potential delay changes caused by the uncertainty of Android audio delay time, cross
The correlation analysis between the target ERP global field power values of the morning and afternoon oddball sessions was performed, and the lag difference was eliminated by moving all event markers based on the identified lag.
The average lag time compensated by this program is 15 MS (
Cross-subject range: 0 ~ 46 u2009 ms).
In order to further ensure that the remaining time uncertainty not considered in this procedure does not affect the ERP delay, no analysis is carried out, and by averaging the large continuous intervals of the width of the 100 ms, the amplitude data of the analysis is 0 u2009 ms.
For ERP analysis, there are still 79 target trials and 419 standard trials per session.
The number of trials in the morning and afternoon was not significant (
Standard: t = 1 =. 16, u2009=u2009. 28; targets: tu2009=u20091. 67, u2009=u2009. 13). Single-
Use regular linear discrimination analysis (LDA)
Classification implemented in BCILAB.
Filter out continuous data from 0. 1 to 6u2009Hz.
The era was extracted from 0 to 800 ms and the era was extracted using artifacts (
Identified as described above)were discarded.
Number of trials per class (
Target, standard)
Balance, leading to a level of 50% chance.
The feature space used includes five non-
The average window is 100 continuous ms (
MS from 200 to 700)
And 16 channels, generating 80 features.
Reduce the risk of excessive use
Shrink LDA using fit with default settings implemented in BCILAB.
Specifically, the data in the morning is trained by a linear classifier and the data in the afternoon is evaluated.
To obtain an effective performance estimate for calibration data, 5-
Folding time Cross
The verification re-sampling procedure was performed.
Percentage of correct classification tests (% accuracy)
As a measure of performance.
Then, the influence of the number of channels on the classification accuracy is discussed through the iterative program.
For this reason, the judgment value of each electrode is calculated by point double, and the channel is arranged in descending order
Serial Association.
For each of the 16 iterations, discard the minimum discrimination channel, repeat the classifier training for the morning data, and evaluate the model obtained on the afternoon data.
Due to the channel selection of the morning data, the model evaluation of the afternoon data is carried out, so the loop problem is avoided.
In order to count the difference between the EEG power obtained from the open and closed eye conditions, the EEG band is defined (
1-3Hz, 4-7Hz, 8-12Hz, 13-30Hz, 3-80 hz)
And the log with power value submitted to the paired t-tests.
As this leads to a large number of statistical tests (
16 electrodes × 5 Frequency bands)
Bonferroni correction is applied (. 05/80)
And only if the effect is below p.
000625 is considered important.
Oddball ERP analysis using 2 × 2 repeated measurement variance analysis (ANOVA)
Including the elements meeting (
Morning, afternoon)and Condition (
Target, standard).
To explore whether the ERP data obtained with cEEGrid contains meaningful spatial and temporal features, the same ANOVA model is repeatedly applied for a single-
Trial classification.
Bonferroni correction for multiple comparisons is applied again (
16 channels × 5 time boxes; pu2009=u2009. 000625).
Further statistical assessment includes testing
Re-test reliability analysis of target ERPs by Pearson correlation, and calculate the target and standard ERP condition impact size by dividing the average difference by the standard deviation of the collection (Cohen’s d).
Test whether the classification accuracy is higher than the opportunity-
The confidence limit is the-level binomial statistics of the value = value 0.
05 was used to assess whether significant loss of classification accuracy occurred between morning and afternoon
Test was applied.