Validation of EntenThey compared the performance of Enten to a Research Grade EEG Headset from cogniomics, called Quick20
They measured the electrical signals in three dfiferent experimetns with known neural responses.
If Enten matched Quick 20s “Research Grade” responses, it serves as validity that their EEG system works at research grade quality.
1st Experiment targets the visual cortex, recording EEG data while the individual closes their eyes.
Expectation was to see increase in alpha activity (8-13 Hz, relaxed conscious regular state)
2nd Experiment targets the auditory cortex, recording EEG data in response to auditory stimulus (sound).
Electrical Activity would increase based on the modulation of the stimulus (sound).
3rd Experiment, Oddball Paradigm
oddball paradigm is a commonly used task for cognitive and attention measurement. stimuli is presented that’s occasionally interrupted by a random oddball stimulus.
A P300 response was expected.
Enten’s Impedance was measured over time to make sure that a good level of skin coupling could be achieved and that hair didn’t interfere with the performance of Enten’s EEG Electrodes.
1st Experiment | visual cortext
The eye closed state would evoke more alpha waves than eyes opened.
Eyes opened would induce a reduction in alpha waves | Berger Effect.
30 seconds eyes opened. 30 seconds eyes closed.
They then conducted a power spectogram analysis to examine the difference in brain wave activity between open and closed on alpha wave patterns.
Quick 20
O1 and O2 electrodes were on the occipital lobes
T3 and T4 electroes were on the temporal lobes.
Figure 2 showed that there were high levels of alpha oscillations (as expected per Berger effect). Enten was able to measure these alpha oscillations similar to Quick20.
Figure 3 showed that PSD of alpha oscillations that were recorded through the temporal lobe were detected at similar levels on enten and quick 20. enten might’ve even been slightly better.
2nd Experiment | Auditory Cortex
ASSR - evoked electrical potentials due to modulated frequency of tones (just noise bro)
Expected to see increases in frequency of the EEG signal to the specific frequency modulation of the tones in the auditory cortex
They played a 1000Hz tone that was modulated at 37Hz and 43Hz at 70% of computer volume.
The 1000Hz tone was turnt on and off 37 times per second
The 1000Hz tone was turnt on and off 43 times per second.
In Enten, there was a peak in brain wave activity of 14dB at 37Hz when the the tone was modulated to 37Hz
There was also a peak in brain wave activity of 7dB at 43Hz when the tone was modulated to 43Hz.
Enten seems to capture those signals with higher sensitivity than Quick 20. Quick 20 wasn’t able to detect with that power through the temporal lobe / auditory cortex channels.
3rd Experiment | P300
Participant was exposed with 4 words. blue, yellow, green, red. They would only press a button when the word blue appears on screen.
The word blue and the button press would elicit a P300 response in brain activity. The P300 response had a very similar amplitude when compared to Quick20 at the temporal lobe.
Impedance
Good skin coupling reached good levels of <300 kilo ohms after a settling time of minutes.
After 10 minutes, about 13-14 channels had <100 kilo ohms. 17 channels after 16 minutes of wear.
Experiments measuring behavior in response to distraction.Neurable developed / designed cognitive experiments to let them analyze changes in focus due to distractions
Stroop Task Experiment
Congruent section yields more accuracy and lowered response time for the person
Incongruent section yields less accuracy and increased response time for the person
This is the stroop effect.
Additional sensory information in audio and video was played to distract the person.
Hypothesis was that the stroop effect would be more pronounced which would represent distractions in daily life settings.
Distractful stimuli were
They provide different levels of distraction.
4 blocks of stroop task w each block containing 64 trials were performed
Equal numbers of congruent / incongruent trial.
Each individual block was associated w a different stimuli scene.
Participants had 1.2 seconds to respond to the specific trial. If exceeded, trial was marked as missed and next trial appeared.
64 unique participants and 80 sessions were recorded using Enten.
There was a drop in alpha band power during the incongruent trials.
To see in what conditions a higher drop would be expected, instead of putting the scene in the backround, the scene was directly in the way of the user when performing incongruent trials.
There was a bigger difference between congruent tasks with no scene and incongruent tasks with a distracting scene.
Interruption by notification
experiment 2a - same thing as stroop test with scene directly impeding but with the addition of a notification and response to chatbot.
It was found that there was more difficulty in accuracy and slower reaction times.
They questioned the role of motion artificats in enten that could potentially influence data. Perhaps the motion artifacts are what incorrectly indicate an unfocused state via electrical activity rather than the brain waves themselves.
So they adjusted experiments 2b and 2c to limit motion artifiacts.
Experiment 2b - stroop test, no scene. notification → mental math problem
Experiment 2c - stroop test, no scene. notification → recall object of same color
Neural activity in response to distraction
neural activity in the alpha band can indicate that we’re filtering out and ignoring distractions. beta activity is also associated with attention.
incongruence in stroop trials were considered distractions.
power in alpha activity was compared between congruent and incongruent trials.
alpha power is lower in incongruent trials (distraction) than congruent trials.
there was also a similar reduction in low beta bands, (13-20 Hz)
They hypothesized that power in alpha (and low beta) would be low during experiment 1b’s hard blocks where scenes were directly impeding the ability of the user to go about the stroop test.
Results confirmed their hypothesis, there was lowered alpha and beta. Incogruent trials (distractions) would lower alpha and low beta activity which then indicates lowered focus.
All this was captured by Enten showing that Enten can be reliable for focus. Alpha band is highly associated with focus.
ML to detect distractions in EEGSupport Vector Machine Classifier was trained to predict if a user was doing an easy or a hard task.
An easy task would be indicative from a good focus level. A hard task would be indicative from a lowered focus level.
It used the power of EEG alpha waves as features.
Features were scaled by subtracting the median value and dividing by the interquartile range (the difference between the first and third quartile)
This allows for large outliers of data from the dataset not to be taken into account in the scaling feature.
Outleirs such as artificats from motion or external devices can come into play.
This method of scaling allows for them to be denied way into the dataset.
It allows for data of one session to be comparable to data from another session.
The probability that a user was in an easy task would be a proxy for focus levels.
Then to test the model, they hypothesized that the distractions of the experiments 2A-C (ones which distractions were notifications), made it difficult to focus | analogous to the hard blocks in experiments 1B (where the busy market place was put in direct impedance of the stroop test)
So they tested the model on Experiments 2A-C, and scored their focus. The score was derived from the neural activity in the alpha band.
Algorithm Robustness to ArtifactsThe algorithm designed to detect and score focus needed to be robust against artifacts.
They found that the drop in focuss in focus, is sustained after distraction meaning it gets harder to refocus once one is already distracted.
Performance in Daily UseTo validate use of the model and consistency + repeatability in iding distractions, they analyzed the singular participant who went in experiment 2A over 3 days. The focus measure was found to drop consistently as a response to distraction over the 3 days. QA scores averaged mid 80% indicating consistency and reliability in daily use.
Then they did a group analysis on 6 subjects yielding simlar results of aobut 80% in QA.
DiscussionNeurable Enten
They validated that their extracted alpha features correlated w focus during stroop test. Meaning alpha band indicates focus level. They validated that distractions can cause lack of focus which can be detected via alpha band in EEG. They validated that Neurable Enten’s electrical data isn’t from EOG or EMG but rather pure EEG from the brain.
ML using features from Alpha band at Temporal lobe is ideal for determing focus state.