Brain signals or artifacts?
Checking whether low-cost EEG systems are really decoding the brain, or picking up eye movements, facial muscles and head motion instead.
2026–27In progressEEGEMGML
The question
When a cheap EEG system classifies motor imagery well, is it reading brain activity, or is it picking up artifacts like eye movements, facial muscle activity and head motion?
The plan
- Record EEG alongside EOG (eye movements), facial EMG and head motion
- Compare EEG-only, artifact-only and combined models
- Repeat the evaluation across recording sessions
Example
- Run a motor imagery classifier.
- Test whether facial EMG alone can predict the class.
- Remove frontal electrodes and contaminated trials.
- Check whether the claimed EEG performance still holds.
What it produces
- Artifact-only baselines
- Channel and frequency ablations
- Same-session versus cross-session results
- A validation checklist for future club projects
Background: a review of EEG artifacts and how to handle them.
Possible alternative: how far can low-cost EEG and EMG go?
This project may be swapped for a related question: what’s roughly the best performance we can get on a given task with our own EEG and EMG hardware?
- Compare three common BCI paradigms (SSVEP, P300 and motor imagery) using 1, 2, 4 and 8 channels
- Limit per-person calibration to 0, 1, 5 and 10 minutes
- Compare classical methods against small neural networks
That version would use public BCI datasets, such as those available through the MOABB benchmark.