This dataset contains EEG, MEG, functional MRI and structural MRI data on the same subject with the same paradigm, which allows a basic comparison of faces versus scrambled faces.
It can be used to demonstrate, for example, 3D source reconstruction of various electrophysiological measures of face perception, such as the ``N170'' evoked response (ERP) recorded with EEG, or the analogous ``M170'' evoked field (ERF) recorded with MEG. These localisations are informed by the anatomy of the brain (from the structural MRI) and possibly by functional activation in the same paradigm (from the functional MRI).
The demonstration below involves localising the N170 using a distributed source method (called an ``imaging'' solution in SPM). The data can also be used to explore further effects, e.g. induced effects (Friston et al, 2006), effects at different latencies, or the effects of adding fMRI constraints on the localisation.
The EEG data were acquired on a 128 channel ActiveTwo system; the MEG data were acquired on a 275 channel CTF/VSM system; the sMRI data were acquired using a phased-array headcoil on a Siemens Sonata 1.5T; the fMRI data were acquired using a gradient-echo EPI sequence on the Sonata. The dataset also includes data from a Polhemus digitizer, which are used to coregister the EEG and the MEG data with the structural MRI.
Some related analyses of these data are reported in Henson et al (2005a, 2005b, 2007, 2009a, 2009b, in press), Kiebel and Friston (2004) and Friston et al (2006). To proceed with the data analysis, first download the data set from the SPM website\footnote{Multimodal face-evoked dataset: \url{http://www.fil.ion.ucl.ac.uk/spm/data/mmfaces/}}.
Most of the analysis below can be implemented in \matlab\ 7.1 (R14SP3) and above. However, recoding condition labels using the GUI requires features of SPM8 only available in \matlab\ 7.4 (R2007a) and above.
\section{Paradigm and Data}
The basic paradigm involves randomised presentation of at least 86 faces and 86 scrambled faces (Figure~\ref{multimodal:fig:1}), based on Phase 1 of a previous study by Henson et al (2003). The scrambled faces were created by 2D Fourier transformation, random phase permutation, inverse transformation and outline-masking of each face. Thus faces and scrambled faces are closely matched for low-level visual properties such as spatial frequency content. Half the faces were famous, but this factor is collapsed in the current analyses. Each face required a four-way, left-right symmetry judgment (mean RTs over a second; judgments roughly orthogonal to conditions; reasons for this task are explained in Henson et al, 2003). The subject was instructed not to blink while the fixation cross was present on the screen.
\caption{\em One trial in the experiment. Trials involved either a Face (F) or Scrambled face (S). \label{multimodal:fig:1}}
\end{center}
\end{figure}
\subsection{Structural MRI}
The T1-weighted structural MRI of a young male was acquired on a 1.5T Siemens Sonata via an MDEFT sequence with resolution $1 \times 1 \times 1 mm^3$ voxels, using a whole-body coil for RF transmission and an 8-element phased array head coil for signal reception.
The images are in NIfTI format in the sMRI sub-directory, consisting of two files:
\begin{verbatim}
sMRI/sMRI.img
sMRI/sMRI.hdr
\end{verbatim}
The structural was manually positioned to roughly match Talairach space, with the origin close to the Anterior Commissure.
The approximate position of 3 fiducials within this MRI space - the nasion, and the left and right peri-aricular points - are stored in the file:
\begin{verbatim}
sMRI/smri_fid.txt
\end{verbatim}
These were identified manually (based on anatomy) and are used to define the MRI space relative to the EEG and MEG spaces, which need to be coregistered (see below). It doesn't matter that the positions are approximate, because more precise coregistration is implemented via digitised surfaces of the scalp (``head shape functions'') that were created using the Polhemus 3D digitizer.
\subsection{EEG data}
The EEG data were acquired on a 128-channel ActiveTwo system, sampled at 2048 Hz, plus electrodes on left earlobe, right earlobe, and two bipolar channels to measure HEOG and VEOG. The 128 scalp channels are named: 32 A (Back), 32 B (Right), 32 C (Front) and 32 D (Left). The data acquired in two runs of the protocol are contained in two Biosemi raw data files:
\begin{verbatim}
EEG/faces_run1.bdf
EEG/faces_run2.bdf
\end{verbatim}
The EEG directory also contains the following files:
\begin{verbatim}
EEG/condition_labels.txt
\end{verbatim}
This text file contains a list of condition labels in the same order as the trials appear in the two files - ``faces'' for presentation of faces and ``scrambled'' for presentation of scrambled faces.
The EEG directory also contains the following files:
\begin{verbatim}
EEG/electrode_locations_and_headshape.sfp
\end{verbatim}
This ASCII file contains electrode locations, fiducials and headshape points measured with Polhemus digitizer.
The 3 fiducial markers were placed approximately on the nasion and preauricular points and digitised by the Polhemus digitizer. Later, we will coregister the fiducial points and the head shape to map the electrode positions in the ``Polhemus space'' to the ``MRI space''.
Also included as reference are some SPM batch files and SPM scripts (though these are recreated as part of the demo):
\begin{verbatim}
EEG/batch_eeg_XYTstats.mat
EEG/batch_eeg_artefact.mat
EEG/eeg_preprocess.m
EEG/faces_eeg_montage.m
\end{verbatim}
\subsection{MEG data}
The MEG data were acquired on a 275 channel CTF/VSM system, using second-order axial gradiometers and synthetic third gradient for denoising and sampled at 480 Hz. There are actually 274 MEG channels in this dataset since the system it was recorded on had one faulty sensor. Two runs (sessions) of the protocol have been saved in two CTF datasets (each one is a directory with multiple files)
\begin{verbatim}
MEG/SPM_CTF_MEG_example_faces1_3D.ds
MEG/SPM_CTF_MEG_example_faces2_3D.ds
\end{verbatim}
The MEG data also contains a \texttt{headshape.mat} file, containing the headshape recorded during the MEG experiment with a Polhemus digitizer.
The locations of the 3 fiducials in the \texttt{headshape.mat} file are the same as the positions of 3 ``locator coils'' the locations of which are measured by the CTF machine, and used to define the coordinates (in ``CTF space'') for the location of the 274 sensors.
Also included as reference are two SPM batch files and two trial definition files (though these are recreated as part of the demo):
\begin{verbatim}
MEG/batch_meg_preprocess.mat
MEG/batch_meg_TFstats.mat
MEG/trials_run1.mat
MEG/trials_run2.mat
\end{verbatim}
\subsection{fMRI data}
The fMRI data were acquired using a gradient-echo EPI sequence on a 3T Siemens TIM Trio, with 32, 3mm slices (skip 0.75mm) of $3\times 3 mm^2$ pixels, acquired in a sequential descending order with a TR of 2s. There are 390 images in each of the two ``Session'' sub-directories (5 initial dummy scans have been removed), each consisting of a NIfTI image and header file:
\begin{verbatim}
fMRI/Session1/fM*.{hdr,img}
fMRI/Session2/fM*.{hdr,img}
\end{verbatim}
Also provided are the onsets of faces and scrambled faces (in units of scans) in the \matlab\ file:
\begin{verbatim}
fMRI/trials_ses1.mat
fMRI/trials_ses2.mat
\end{verbatim}
and two example SPM batch files (see Section~\ref{multimodal:data:fMRI}):
\begin{verbatim}
fMRI/batch_fmri_preproc.mat
fMRI/batch_fmri_stats.mat
\end{verbatim}
\section{Getting Started}
You need to start SPM8 and toggle ``EEG'' as the modality (bottom-right of SPM main window), or start SPM8 with \texttt{spm eeg}. In order for this to work you need to ensure that the main SPM directory is on your \matlab\ path.
\input{multimodal/eeg}
\input{multimodal/meg}
\input{multimodal/fmri}
\input{multimodal/fusion}
\section{References}
\noindent 1. Friston, K, Daunizeau, J, Kiebel, S, Phillips, C, Trujillo-Barreto, N, Henson, R, Flandin, G, Mattout, J (2008). Multiple sparse priors for the M/EEG inverse problem. Neuroimage, 39(3):1104-20.\\
\noindent 2. Friston, K, Carlton Chu, Janaina Mouro-Miranda, Oliver Hulme, Geraint Rees, Will Penny and John Ashburner (2008). Bayesian decoding of brain images. NeuroImage, 39(1):181-205.\\
\noindent 3. Friston K, Henson R, Phillips C, and Mattout J. (2006). Bayesian estimation of evoked and induced responses. Human Brain Mapping, 27, 722-735.\\
\noindent 4. Henson, R, Goshen-Gottstein, Y, Ganel, T, Otten, L, Quayle, A. and Rugg, M. (2003). Electrophysiological and hemodynamic correlates of face perception, recognition and priming. Cerebral Cortex, 13, 793-805.\\
\noindent 5. Henson R, Mattout J, Friston K, Hassel S, Hillebrand A, Barnes G and Singh K. (2005a) Distributed source localisation of the M170 using multiple constraints. HBM05 Abstract.\\
\noindent 6. Henson R, Kiebel S, Kilner J, Friston K, Hillebrand A, Barnes G and Singh K. (2005b) Time-frequency SPMs for MEG data on face perception: Power changes and phase-locking. HBM05 Abstract.\\
\noindent 7. Henson, R, Mattout, J, Singh, K, Barnes, G, Hillebrand, A and Friston, K.J. (2007). Population-level inferences for distributed MEG source localisation under multiple constraints: Application to face-evoked fields. Neuroimage, 38, 422-438.\\
\noindent 8. Henson, R, Mattout, J, Phillips, C and Friston, K.J. (2009a). Selecting forward models for MEG source-reconstruction using model-evidence. Neuroimage, 46, 168-176.\\
\noindent 9. Henson, R, Mouchlianitis, E and Friston, K.J. (2009b). MEG and EEG data fusion: Simultaneous localisation of face-evoked responses. Neuroimage, 47, 581-589.\\
\noindent 10. Henson, R, Flandin, G, Friston, K.J. and Mattout, J. (in press). A Parametric Empirical Bayesian framework for fMRI-constrained MEG/EEG source reconstruction. Human Brain Mapping.\\
\noindent 11. Kilner, J., Kiebel, S and Friston, K. J. (2005). Applications of random field theory to electrophysiology. Neuroscience Letters, 374:174-178.\\
\noindent 12. Kilner, J. and Penny. W. (2006). Robust Averaging for EEG/MEG. HBM06 Abstract.\\
\noindent 13. Kiebel S and Friston K (2004). Statistical Parametric Mapping for Event-Related Potentials II: A Hierarchical Temporal Model. NeuroImage, 22, 503-520.\\
\noindent 14. Kiebel, S.J., David, O. and Friston, K. J. (2006). Dynamic Causal Modelling of Evoked Responses in EEG/MEG with lead-field parameterization. NeuroImage, 30:1273-1284.\\
\noindent 15. Mattout J, Pelegrini-Issac M, Garnero L and Benali H. (2005a). Multivariate source prelocalization (MSP): use of functionally informed basis functions for better conditioning the MEG inverse problem. Neuroimage, 26, 356-73.\\
\noindent 16. Mattout, J, Phillips, C, Penny, W, Rugg, M and Friston, KJ (2005b). MEG source localisation under multiple constraints: an extended Bayesian framework. NeuroImage.\\
\noindent 17. Mattout, J., Henson, R N. and Friston, K.J. (in press). Canonical Source Reconstruction for MEG. Computational Intelligence and Neuroscience.\\
\noindent 18. Phillips, C, M.D. Rugg, M and Friston, K.J. (2002). Systematic Regularization of Linear Inverse Solutions of the EEG Source Localisation Problem. NeuroImage, 17, 287-301.\\
\noindent 19. Spinelli, L, Gonzalez S, Lantz, G, Seeck, M and Michel, C. (2000). Electromagnetic inverse solutions in anatomically constrained spherical head models. Brain Topography, 13, 2.\\
\noindent 20. Wager, TD, Keller, MC, Lacey SC and Jonides J. (2005). Increased sensitivity in neuroimaging analyses using robust regression. NeuroImage, 26(1), 99-113.\\