Dealing with artifacts

[SveaMeyer13]

There are different artifacts that occur in EEG data. They include environmental artifacts such as the AC powerline frequency which can be removed via filtering. But also artifacts that stem from biological sources:

• sweat leads to electrode drift
• eye movements lead to strong signal deflections (mostly frontal channels)
• burst due to swallowing or clenching teeth
• flatlines due to lost signal
• movement artifacts
• heartbeat
  ...

Different ways of detecting and rejecting or repairing such data exist. Many of them however require additional information other than just the time series information we currently hold in sktime (simply timepoints * channels) such as information from eog or ecg channels (for automated repair based on ICA), labeled parts of clean data for Artifact Subspace Reconstruction or information on exact channel locations for RANSAC.

A more thorough review of methods and algorithms for artifact removal would be required to identify what makes sense for sktime-neuro.

A few pointers:

• https://eeglab.org/tutorials/06_RejectArtifacts/
• http://vislab.github.io/EEG-Clean-Tools/
• https://github.com/autoreject/autoreject

[fkiraly]

Hm, this sounds quite complex.

I suggest you take a step back to the drawing board and try to identify what the scientific types of these operations are - what goes in, what comes out, what does it do?

I'm unsure myself from glancing only quickly at the references, but I think I can see:

• labelled variables, i.e., variables that have specific meaning
• much of this seems to be annotation and annotation conditional removal?