How many projectors should I use in the SSP (Signal Space Projection) method for EEG preprocessing ?

I want to preprocess 21 Channels of EEG ( and 3 more channels for EOG) data using MNE. In order to remove EOG artifacts, I use SSP in MNE (Python).

When I increase the number of projectors, the SSP removes EOG artifacts.

But I think besides removing the artifacts I lose some information in raw data.

Based on a tutorial in the MNE toolbox:

How did we choose these numbers (number of projectors)? The short answer is “based on experience”. ([Repairing artifacts with SSP — MNE 1.0.0 documentation]

But I couldn’t find any article regarding this problem, I mean finding the optimum number of projectors.

Hello @Borjkhani and welcome to the forum!

MNE defaults to 2 projectors per channel type and artifact, but in practice when working with @agramfort we normally only add a single projector per channel type for each artifact. This is usually sufficient to remove EOG and ECG artifacts.

How do you create your projector(s)?

You are correct in assuming that adding more projectors is also more likely to remove physiological brain data.

Best wishes,
Richard

For EOG/ECG artifacts, you could also use an ICA decomposition: tutorial. The components capture very well the ocular and heartbeat activity.

1 Like