|
Size: 2135
Comment:
|
← Revision 8 as of 2013-03-08 10:02:45 ⇥
Size: 2129
Comment: converted to 1.6 markup
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 8: | Line 8: |
| 1. Create forward models by calling Brainstorm (concentric spheres, overlapping-spheres, BEMs) Invert forward models using Multiple Sparse Priors (Friston et al, 2008), an important new approach that uses ~750 local cortical patches as source priors… … including ability to optimise source priors by pooling over subjects (Litvak & Friston, 2008)… … and (in near future) the ability to add multiple fMRI-cluster source priors (Flandin et al, in prep) 1. Then simultaneously re-invert (fuse) forward models for each sensor-type, which automatically weights each sensor-type (magnetometers, gradiometers, EEG) in a principled fashion (Henson et al, submitted) 1. (Compare different models using the Bayesian model evidence, eg Henson et al, in press) |
1. Create forward models by calling Brainstorm (concentric spheres, overlapping-spheres, BEMs) 1. Invert forward models using Multiple Sparse Priors (Friston et al, 2008), an important new approach that uses ~750 local cortical patches as source priors… … including ability to optimise source priors by pooling over subjects (Litvak & Friston, 2008)… and (in near future) the ability to add multiple fMRI-cluster source priors (Henson et al, submitted) 1. Then simultaneously re-invert (fuse) forward models for each sensor-type, which automatically weights each sensor-type (magnetometers, gradiometers, EEG) in a principled fashion (Henson et al, 2009b) 1. (Compare different models using the Bayesian model evidence, eg Henson et al, 2009a) |
Outline analysis pipeline for Neuromag MEG/EEG data in SPM
- Use Neuromag's Maxfilter (including Trans Default for Sensor-Level analyses)
- Write magnetometers, gradiometers and EEG to separate files for parallel preprocessing
- Call EEGLAB from SPM to project out ICA components that correlate with measured EOG/ECG
- Usual filtering, epoching, thresholding, averaging…
- Write out 2D sensor x 1D time images for each trial or subject, and localise reliable voxel-wise effects in 3D space-time across trials/subjects using Random Field Theory for multiple comparisons
- Automatically normalise and segment MRI, and create meshes for cortex, skull and scalp (“canonical” cortical mesh is an inverse-normalised template mesh, Mattout et al, 2007)
- Create forward models by calling Brainstorm (concentric spheres, overlapping-spheres, BEMs)
- Invert forward models using Multiple Sparse Priors (Friston et al, 2008), an important new approach that uses ~750 local cortical patches as source priors…
… including ability to optimise source priors by pooling over subjects (Litvak & Friston, 2008)… and (in near future) the ability to add multiple fMRI-cluster source priors (Henson et al, submitted)
- Then simultaneously re-invert (fuse) forward models for each sensor-type, which automatically weights each sensor-type (magnetometers, gradiometers, EEG) in a principled fashion (Henson et al, 2009b)
- (Compare different models using the Bayesian model evidence, eg Henson et al, 2009a)
- Evaluate a time-frequency contrast of source energy (which can include induced energy, Friston et al, 2006)
- Write a 3D image in template (MNI) space for that contrast, and perform usual SPM voxel-wise analysis across trials/subejcts to localise effects in the brain
- (A similar pathway can also be used for time-frequency analysis using wavelets)
- A further possibility is to assume a number of ECDs (eg seeded by distributed inversion above), and use Dynamic Causal Modelling (DCM) to make inferences about changes in effective connectivity …[more will follow on DCM]