Hands On With EEG Data

Starting interactive Python…

Initializing...

1. Raw

Try draggin the selectro strip to see the data! for ease we have gain set lower, use +/- to see various configs. Here the shaded region is eyes closed, and the ticks are blink beeps. Channels here swing up to 21,118 µV. Alpha is 10–50 µV.

2. Filtering frequencies

The first tool in the signal processing toolbox to try is always the frequency filter, either low pass or hihg pass. given the alpha frequency we are trying to find, what do you think would be the best signals to keep?

The slow drift is still in. The 60 Hz hum is still 7× its neighbouring frequencies.

3. Broken electrodes

As we remember from the in class demo, there is a very good chance that a few if note many of the electrodes are not fitted properly or broken. Especia.ly true for dry electrode methods! This graph shows an electrodes spread compared to the others, which should help us identify outliers. For cleaning the data, click the ones you think are broken/unseated! Theres also an automatic detector which you can configure by dragging the line.

4. Reference

We choose a reference here as a value we substract from each electrode channel. Below, the electrodes are colored by how much the alpha signal grows when the eyes are closed. if you hover you can see exact values.

5. ICA ( Independent component analisys)

ICA is a technique that seperates components of the full signal we see into the features that create it. This helps us to isolate what sources of signal are contributing to the alpha waves we seek to extract, and what may just be blinks. each of these cards represents one of these components, the 10-20 System diagram showing the elctrodes that contribute, and a signal sample of 10s of blinks + spectrum. Try to select/deselect components that help . Below, faint is before and ink is after.

Component 16 is the blink and is still in. Best here: 3.9× alpha, 125 µV of blink; yours: 4.7×, 228 µV.

6. Rejection

Here we want to try and find the best rejection boundary that cuts off all blink peaks on a given channel, while minimizing the amount of actual signal we are cutting out. If you have done anything with logistic models or F1 score work you would know this as choosing our accuracy/precision tradeoff. Try to drag the line to keep as much of the eyes closed as possible while getting rid of most if not all of the blinks.

The final signal

Here you can now see the result of your processing, or even the best possible processing given parameter options here. This is important to note because of how much more difficult it becomes to reasonably isolate this signal with the poorer EEG headsets. you might have noticed in the prior sections how gel cap EEG required no sensor dropout. Yours is what you did above; best is the best this pipeline gets from each headset. Alpha is eyes-closed over eyes-open 8–13 Hz power at the back channel; blink is the eyes-open voltage at the front (95th percentile). Turn on events to check what you saw.