Gain matrix workflow

[maedoc]

Gain matrices map source space (neural mass state variables) to sensor space (MEG, EEG and sEEG recordings). They are required for simulating MEG, EEG and sEEG; they are also required for fitting data with those modalities. While most software assumes one single gain matrix, in cases where we have multiple source spaces and sensors spaces (cortical + subcortical; MEG & EEG), then we are computing M * N gain matrices.

Our current scripts on these topics don't work, but they outline the necessary steps, using the OpenMEEG command line utilities. OpenMEEG works primarily with text files for describing head geometry, interface conductance, source geometry & sensor geometry, and MATLAB format data files.

We can certainly format all our data to such files, but note that we want to work with corresponding time series data: we should consider using the MNE-Python library for this, as MNE data structures coordinate time series data with geometry information about source and sensor spaces (though MNE's use of OpenMEEG needs to be built). In general, for any sort of time series we have, it'd be good to build on MNE instead of recoding (badly) their functionality.

Our workflow looks like

• loading time series datasets
• extracting geometry information
• aligning with T1 coord sys (has to be done manually, but MNE has some visual tools for this)
• building the head, sensor, source head models
• inverting head model
• computing gain matrix per (source space, sensor space) pair
• saving for later use