Epochs Normalization and Frontal Asymmetry

Hi Mike, I am in my master’s project, I am making a system that detects the stress level of drone pilots in real time. To do this, I previously generated a database.

I have seen in several sources where it refers to frontal asymmetry (mainly with alpha rhythm)
to detect stress and other emotional states. Do you have videos or is there a section of your books that addresses this topic?

In addition, the interest of the project is to detect stress using Deep Learning. So I am generating the scalograms with CWT. I have wanted to normalize the signals by epoch, but I run into the problem that I lose the relationship, for example, if I normalize from 0 to 1, when concatenating the epoch the signal looks very flat. Is there a way to normalize the signals without losing the relationship between each epoch?
I saw your video where you use log(activity/baseline), but I can’t find how to calculate the baseline for my problem.

Hi Daniel. I don’t have any videos about frontal alpha asymmetry in a lot of detail, but the methods aren’t too difficult – you just compute alpha power at two fontal electrodes and take their log ratio. Perhaps this paper will be helpful (I’m sure there are many many other good methods-oriented asymmetric papers, but I happen to know about this one).

As for normalization… that’s always tricky business. If you have resting state data, there is no real “baseline.” Perhaps one idea is to fit and remove the background 1/f of the power spectrum. This can be done as simply as fitting a line through the power spectrum in log-log space (that is, log-frequency by log-power). There are also more advanced methods, e.g., FOOOF.

1 Like

Thanks so much for the great advice. I will look for information about it, I had never read about FOOOF. If I have the solution in a couple of days, I will notify it in case it helps someone else.