Learning to repair artifacts in brain signals

Electric and magnetic brain signals recordings (M/EEG, sEEG, LFP) are prone to a contamination by so-called artifacts, which do not have a neural origin but are caused by sensor problems, external electro-magnetic sources, or by bio-signals such as muscle activity. Artifacts can mask the neural signal to an extent, that its interpretation by humans or signal analysis pipelines becomes more difficult or even renders it impossible. Besides removing contaminated time intervals of the recording or strongly affected channels, cleaning procedures based on independent component analysis (ICA) have been proposed in the past. This project investigates if deep learning methods can be trained to reliably correct such artifacts in EEG recordings.

Skills required

• Machine learning
• Good programming skills in Python (familiarity with numpy, sklearn, at least one popular deep learning framework).
• Good mathematical background and intuition.