Free software solutions for MEG/EEG source imaging

Free software solutions for MEG/EEG source imaging François Tadel Cognitive Neuroscience & Brain Imaging Lab., CNRS University of Paris - Hôpital de la Salpêtrière Cognitive Neuroimaging Unit, Inserm U562 NeuroSpin - CEA France

I. Source imaging steps II. III. Software classification BrainStorm demo

Functional (EEG/MEG) I. Source imaging steps Pre-processing (EEG/MEG) Source imaging Post-processing Pre-processing (MRI) Anatomy (MRI) Interpretation

Pre-processing (EEG/MEG) Basic but important processes: Artifact detection, rejection or reduction Noise reduction Bandpass filtering Averaging Software solutions: Commercial packages (from MEG/EEG vendors or 3rd-party) Academic (free, open-source) solutions: EEGLab, FieldTrip, MNE, Cartool, BrainStorm, etc.

Pre-processing (MRI) Operations: Importation Segmentation of tissue envelopes scalp, skull, cortex (white vs. grey matter), cerebellum Surface meshes Dedicated academic software packages: BrainVISA / Anatomist, FreeSurfer, BrainSuite, SPM, etc.

Source imaging Processes: Co-registration IRM EEG/MEG Forward modelling Inverse estimation Software: BrainStorm MNE EEGlab FieldTrip LORETA NutMEG

Post-processing Operations: Inter-individual co-registration Statistical analysis: exploration and inference Frequency-spectrum analysis and time-frequency decomposition Software: BrainStorm MNE EEGLab FieldTrip LORETA NutMEG

I. Source imaging steps II. Software classification III. BrainStorm demo

II. Software classification Modality : MEG and/or EEG Source model : Dipoles fitting or distributed sources Volume or surfaces Individual or template anatomy Interface: GUI or command line Database oriented or single analysis Interoperability with other systems Software status: Free, open source, commercial

EEGLab MEG/EEG analysis: ERP segmentation, filtering, artifact rejection, averaging, ICA (Independent Component Analysis) Time/frequency analysis Source localization : dipole fitting only Interface: Nice graphical interface for visualization and computation Possible to work on many subjects at the same time Software status: Matlab toolbox / work on all operating systems / Open source Very popular : lots of plug-ins available Import most of the existing file formats Full inter-operability with BESA and FieldTrip

FieldTrip MEG/EEG analysis: ERP segmentation, filtering, artifact rejection, averaging, Advanced time/frequency analysis Source localization : Many forward and inverse models available (distributed and dipolar), on full MRI volume Many statistics and display tools, both on channels and sources Interface: Command line only : difficult to learn Users need to have a good programming experience Software status: Matlab toolbox / work on all operating systems / Open source Import most of the existing file formats Inter-operability with BESA, EEGLab and LORETA

Cartool EEG analysis: ERP segmentation, filtering, artifact rejection, averaging, Advanced time/frequency analysis Many statistical tools ERP segmentation in microstates Source localization : Computed by external programs, and visualized by Cartool Interface: Fast and beautiful MRI, EEG and sources visualization Good interactivity, easy data exploration Software status: Free but not open source, C++, Windows only Not widely distributed, but very stable Import most of the existing file formats

BrainVISA / Anatomist 3D/4D image processing: Automated and semi-automated pipelines for smri and fmri processing Advanced MRI segmentation Extraction of head and cortical surfaces Source localization: Fully automated pipeline for EEG/MEG sources reconstruction Based on BrainStorm algorithms (distributed models, cortically constrained) Interface: Very rich visualization interface (Anatomist) Advanced database system: efficient classification of recordings Adapted to process numerous subjects and conditions But not very interactive for EEG/MEG recordings and sources Software status: Python / work on all operating systems / Open source

BrainSuite Structural MRI : Automated extraction of head and cortical surfaces Extraction of skull (inner and outer surfaces) Results for surface reconstruction are better with BrainVISA. Interface: Minimal visualization of MRI and surfaces Simple and fast 10 minutes to extract a cortex surface when using the software for the first time Software status: C++/MFC, Windows only / Free but not open source Last update: June 2005

FreeSurfer Structural MRI : Advanced MRI segmentation Automated extraction of head and cortical surfaces Good methods for inflation and normalization of cortical surfaces Interface: GUI for surfaces and volumes visualization and manipulations : useful but could be more interactive Computations from command line only. Software status: C / Linux, Solaris, IRIX, MacOSX / Open source Popular software : many users an developers

MNE MEG/EEG analysis: ERP segmentation, filtering, artifact rejection, averaging, Source localization : Many forward models available Minimum norm estimator, cortically constrained Possible to constrain source estimator with fmri Interface: Everything accessible from the command line Some features have a graphical interface Software status: Compiled C, Linux and MacOS/X + Matlab toolbox for reading Neuromag/FIF file format Free but not open source Import most of the existing file formats Linked with FreeSurfer

SPM5 MEG/EEG analysis: ERP segmentation, filtering, artifact rejection, averaging, Statistical analysis: Advanced statistical methods for fmri processing Volume normalization (MNI or Talairach space) for group-based analysis Possibility to process EEG/MEG sources and fmri in the same environment Source localization: Minimum norm estimator, not constrained Sources in full MRI volume Interface: Graphical interface is poor and not interactive Software status: Matlab / work on all operating systems / Open source Very popular : reference for fmri statistical processing

LORETA Source localization: EEG only Minimum Laplacian estimator, not constrained Impossible to use individual anatomy Statistical methods on channels and sources Interface: Simple GUI, poor results Software status: Compiled executable / Windows only Very few compatibility with other systems

BrainStorm Source localization: Many forward models available Many sources estimator available, all cortically constrained (minimum norm, beamformers, RAP-MUSIC) Statistical analysis on both channels and sources Minimalist surfaces normalization, for inter-subjects analysis Interface: Advanced and intuitive user interface for visualization, computation, and database manipulation Very interactive: best tool for data exploration Database oriented : protocols/subjects/ Software status: Matlab / work on all operating systems / Open source Inter-operability: Imports files from EEGLab, FreeSurfer, BrainSuite, BrainVISA, Cartool, BESA

Conclusion Three or four independent steps: Pre-processing of MEG/EEG surface data Pre-processing of MRI (only if using individual anatomy) Source estimation Post-processing of source dynamic maps Possible good combinations of software packages Examples: EEGLab + BrainVISA + BrainStorm (Matlab) Cartool + BrainSuite + BrainStorm + SPM (Windows) FreeSurfer + MNE (Linux) EEGLab + FieldTrip (Matlab)

Commercial packages BESA ASA Curry EMSE BrainVoyager

References EEGLab: Swartz Center for Computational Neuroscience University of California San Diego (US) http://sccn.ucsd.edu/eeglab/ FieldTrip: F.C. Donders Centre for Cognitive Neuroimaging University Nijmegen (NL) http://www2.ru.nl/fcdonders/fieldtrip Cartool: D.Brunet, Functional Brain Mapping Laboratory University of Geneva (CH) http://brainmapping.unige.ch/cartool.htm BrainVISA / Anatomist: Various research groups around Paris (CNRS, INSERM, CEA) http://brainvisa.info

References BrainSuite: DW. Shattuck, University of California Los Angeles and University of Southern California (US) http://brainsuite.usc.edu/ FreeSurfer: Athinoula A. Martinos Center for Biomedical Imaging Massachusetts General Hospital (US) http://surfer.nmr.mgh.harvard.edu MNE: M.Hämäläinen, A.A. Martinos Center for Biomedical Imaging Massachusetts General Hospital (US) www.nmr.mgh.harvard.edu/martinos/userinfo/data/sofmne.php SPM5: Wellcome Trust Centre for Neuroimaging University College London (UK) http://www.fil.ion.ucl.ac.uk/spm

References LORETA: RD Pascual-Marqui, KEY Institute for Brain-Mind Research University of Zürich (CH) http://www.uzh.ch/keyinst/newloreta BrainStorm: University of Southern California (US) Los Alamos National Laboratory (US) CNRS & INSERM, Paris (France) http://neuroimage.usc.edu/brainstorm

I. Source imaging steps II. Softwares classification III. BrainStorm demo

BrainStorm Database MEG/EEG recordings and results from computations are stored in a database Three levels of classification: Protocols Subjects Conditions Subjects data include: Anatomical data: MRI, head tissue envelopes Functional data: recordings, sensors locations, computed sources, statistical analysis

Exploration mode Time window Database explorer Display setup Messages console

Selected protocol Subject 001 Condition VM

View scalp, cortex, electrodes

View EEG recordings

View EEG topography

Compute sources

View cortical sources

Changing threshold and colormap

Brain smoothing

Definition of a ROI

Activation of the ROIs

Contrast between two conditions

Contrast between two conditions