- SSP - Signal Space Projection
- Projects whatever vector you give as input orthogonally to the data set. You can use this to remove empty room readings (essentially a rough way to remove the ambient noise that is always present) or to remove PCA-components that represent artifacts.
- SSS - Signal Space Separation
- Based directly on Maxwell's Equation, this is a method for removing external-to-head noise. Unlike SSP, this removes noise beyond just empty-room, including noise that may be present only during the session you are working on. Can get rid of some external-to-head body artifacts (hopefully cardiac).
- tSSS - Spatio-Temporal SSS
- This is SSS which considers the extra dimension of time. It can get rid of some within-head artifacts.
MaxFilter is an Elekta Neuromag software package that provides various functions for cleaning up data.
Available MaxFilter functions:
- SSS and tSSS (described above)
- Centering head position into a common frame. This can be useful for aligning data between runs and/or subjects.
- Automatic detection and removal of bad channels (note: this does not seem to be working on the CABMSI HP-UNIX machines, though it works with Linux installations, as of 2008-03-05).
- Head movement compensation (only if using continuous HPI acquisition; not currently available at CABMSI).
Caution Using Maxfilter
- By using SSS dimension of data become 64, so normal ICA or algorithms based on data covariances have to be modified to keep this checked.
- tSSS which is Signal projection in temporal domain, based on common space between B(in) and B(residual), it changes covariance matrix at every 4 sec as different component are projected out at every 4 sec (or the value set in GUI), algorithms based on data covariances have to keep this aspect checked.
See Using MaxFilter for tips and guidelines for using the software.
Artifact Detection and Removal
It has been known since last 200 years, accurate estimation of current distribution from magnetic field is not possible due to the non trivality of null space of the lead field matrix that connect current distribution to measured magnetic fields. In order to overcome this some sort restrictive prior are needed for current distribution, so broadly MEG source localization methods can be divided in three parts.
- Dipole Fitting
- Multiple Dipole Fitting
- Extended Patch Fitting (To correctly model surface extent of current distribution)
- MUSIC--MUltiple SIgnal Classification
- RAP MUSIC--Recursively Applied (RAP) MUSIC
- LCMV--Linear Constrained Minimum Variance
- MCE - Minimum Current Estimate
- MNE - Minimum Norm Estimate
- LORETA - Low Resolution Brain Electromagnetic Tomography
- Probably everything you ever wanted to know is at this website by Pascual-Marqui
- dSPM - dynamic Statistical Parametric Mapping
- sLORETA - standardized Low Resolution Brain Electromagnetic Tomography