An automated EEG repair tool Kristjan-Julius Laak
mission Automate the phase of cleaning EEG data from non-brain- related activity. impossible?
Step by step one goes very far (proverb)
Mission possible AUTOMATE ONLY SOME ASPECTS OF THIS PHASE
Who cares at all? 200 recordings .. 10 subjects ... 60 electrodes ....
!!!!!!!!!!!!!?!!!!!!!!!!!!!! 256
Goal Build an automated algorithm that detects and repairs channels containing noise or artifacts.
Prerequisites • Data is already preprocessed (Fieldtrip) • There are trials not a single continuous recording • Some aspects of the data are known (e.g. sample rate)
General pipeline Euclidean Probably ±250µV distance from density median estimate Visual ICA Interpolation rejection
General pipeline Euclidean Probably ±250µV distance from density median estimate Visual ICA Interpolation rejection
Euclidean distance from median
General pipeline Euclidean Probably ±250µV distance from density median estimate Visual ICA Interpolation rejection
Density estimate A. Probability density estimate plots for each channel per trial. B. The red line on is a reflection of the left side curve from the maximum, illustrating a Gauss curve.
General pipeline Euclidean Probably ±250µV distance from density median estimate Visual ICA Interpolation rejection
Bad channels (automated)
General pipeline Euclidean Probably ±250µV distance from density median estimate Visual ICA Interpolation rejection
General pipeline Euclidean Probably ±250µV distance from density median estimate Visual ICA Interpolation rejection
Independent component analysis
Results The difference between raw data (A) and pre-processed data (B) for the same trial. Different colours mark different channels.
Last slide.
Intentionally black slide... Thank you!
Recommend
More recommend