mne group analysis presentation @ biomag 2016 conf

How to perform MEG group analysis with MNE

MNE software for processing MEG and EEG data, A. Gramfort, M. Luessi, E. Larson, D. Engemann, D. Strohmeier, C. Brodbeck, L. Parkkonen, M. Hämäläinen, Neuroimage, 2014MEG and EEG data analysis with MNE-Python, A. Gramfort, M. Luessi, E. Larson, D. Engemann, D. Strohmeier, C. Brodbeck, R. Goj, M. Jas, T. Brooks, L. Parkkonen, M. Hämäläinen, Frontiers in Neuroscience, 2013

Biomag 2016http://www.biomag2016.org/satellite_meetings2.php

http://martinos.org/mne

http://martinos.org/mne

• MNE based on C code developed for ~15 years by MSH• MNE-Python started ~6 years ago at MGH, Boston

About the project

Source: https://www.ohloh.net/p/MNE

https://www.ohloh.net/p/MNE

http://martinos.org/mne/


http://mne-tools.github.io/mne-biomag-group-demo/

People behind this work

@agramfort @dengemann

@jasmainak

http://martinos.org/mne/stable/whats_new.html@jaeilepp

@Eric89GXL

https://github.com/agramfort

https://github.com/dengemann


http://martinos.org/mne/stable/whats_new.html

https://github.com/Eric89GXL

(Some) MNE People

@agramfort @mluessi @dengemann

@lauriparkkonen

@Eric89GXL

@mshamalainen

@t3on

@joewalter

@jasmainak@rgoj

@chris=anmbrodbeck

@jaeilepp

@adykstra @leggi@a

@kazemakase

@TalLinzen

@OlafHauk @jdammers


@kingjr

https://github.com/agramfort

https://github.com/mluessi

https://github.com/mluessi



https://github.com/lauriparkkonen

https://github.com/lauriparkkonen









http://github.com/TalLinzen


MNE People

https://github.com/mne-tools/mne-python/graphs/contributors

Alan Leggitt, Alexander Rudiuk, Alexandre Barachant, Alexandre Gramfort, Andrew Dykstra, Asish Panda, Basile Pinsard, Brad Buran, Camilo Lamus, Cathy Nangini, Chris Holdgraf, Christian Brodbeck, Christoph Dinh, Christopher J. Bailey, Christopher Mullins, Clemens Brunner, Clément Moutard, Dan G. Wakeman, Daniel McCloy, Daniel Strohmeier, Denis A. Engemann, Emanuele Olivetti, Emily Ruzich, Emily Stephen, Eric Larson, Fede Raimondo, Federico Raimondo, Félix Raimundo, Guillaume Dumas, Hafeza Anevar, Hari Bharadwaj, Ingoo Lee, Jaakko Leppakangas, Jair Montoya, Jean-Remi King, Johannes Niediek, Jona Sassenhagen, Jussi Nurminen, Kambiz Tavabi, Keith Doelling, Lorenzo De Santis, Louis Thibault, Luke Bloy, Mads Jensen, Mainak Jas, Manfred Kitzbichler, Manoj Kumar, Marian Dovgialo, Marijn van Vliet, Mark Wronkiewicz, Marmaduke Woodman, Martin Billinger, Martin Luessi, Matt Tucker, Matti Hamalainen, Michael Krause, Mikolaj Magnuski, Natalie Klein, Nick Foti, Nick Ward, Niklas Wilming, Olaf Hauk, Phillip Alday, Praveen Sripad, Richard Höchenberger, Roan LaPlante, Romain Trachel, Roman Goj, Ross Maddox, Sagun Pai, Saket Choudhary, Simon Kornblith, Simon-Shlomo Poil, Sourav Singh, Tal Linzen, Tanay, Teon Brooks, Tom Dupré la Tour, Yaroslav Halchenko, Yousra Bekhti, Ellen Lau, Mads Jensen !

https://github.com/mne-tools/mne-biomag-group-demo/

Get the code to replicate

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/index.html

Analysis Scripts

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/index.html

One script

.../mne-biomag-group-demo/tree/master/scripts/processing

Fetch data from OpenfMRI

Sensor space analysis

Source space analysis

Anatomical pipeline

https://github.com/mne-tools/mne-biomag-group-demo/tree/master/scripts/processing

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/index.html

Results at group level



Demos and results for each

subject


Installing MNE-Python

$ pip install --upgrade --user mne

Install Scientific Python environment, e.g., Anaconda at

http://martinos.org/mne/stable/install_mne_python.html

https://www.continuum.io/downloads

Install MNE-Python:

https://www.continuum.io/downloads





Anatomical pipeline


Built on top of FreeSurfer$ recon-all -s ${SUBJECT} -i xy000000.nii -all

http://surfer.nmr.mgh.harvard.edu/

http://surfer.nmr.mgh.harvard.edu/

Anatomy workflow

MRI data

MRI data reconstructedrecon-all (Freesurfer 5.1)

BEM meshmne watershed_bemor mne flash_bem

BEM model

MEG/EEG data

Forward solutionor gain matrix

not automatic

needs freesurfer

Coregistrationmne_analyze/mrilab/Python

BEM meshes

http://martinos.org/mne/stable/auto_tutorials/plot_forward.html

http://martinos.org/mne/stable/auto_tutorials/plot_forward.html#sphx-glr-auto-tutorials-plot-forward-py

BEM meshes

>>> for subject_id in range(1, 20):>>> subject = "sub%03d" % subject_id>>> mne.bem.make_watershed_bem(subject, >>> subjects_dir=subjects_dir,>>> overwrite=True)

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/11-make_watershed.html

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/11-make_watershed.html

Preprocessing: From raw to ERP/ERFdrifts blinks

line noise

cardiac

10 seconds filtered data clean data using SSPs

Preprocessingto get cleanevoked data(ERF/ERP)

or ICA

Maxwell filtering

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_maxfilter.html

MNE implements maxfilter (TM)

Filtering

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_analysis_1.html


Code for filtering

>>> for subject_id in range(1, 20):>>> subject = "sub%03d" % subject_id>>> for run in range(1, 7):>>> raw_in = ...>>> raw = mne.io.read_raw_fif(raw_in, preload=True, add_eeg_ref=False)>>> raw.filter(1, 40, l_trans_bandwidth=0.5, h_trans_bandwidth='auto',>>> filter_length='auto', phase='zero', fir_window='hann')

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/02-python_filtering.html

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/02-python_filtering.html

Filter design

http://martinos.org/mne/stable/auto_tutorials/plot_background_filtering.html


Filter design

High pass filtering removes slow drifts and can make baselining unnecessary… yet the story is not that simple



Paradigm

Presentation Screen

Famous

Unfamiliar

Scrambled

Paradigm on trigger channel

>>> mask = 4096 + 256 # mask for excluding high order bits>>> events = mne.find_events(raw, stim_channel='STI101',>>> consecutive='increasing', mask=mask,>>> mask_type='not_and', min_duration=0.003)>>> fig = mne.viz.plot_events(events)

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/03-run_extract_events.html

Paradigm in code

>>> events_id = {>>> 'face/famous/first': 5,>>> 'face/famous/immediate': 6,>>> 'face/famous/long': 7,>>> 'face/unfamiliar/first': 13,>>> 'face/unfamiliar/immediate': 14,>>> 'face/unfamiliar/long': 15,>>> 'scrambled/first': 17,>>> 'scrambled/immediate': 18,>>> 'scrambled/long': 19,>>> }

Use “tags” to describe events

From Epochs to Evoked

>>> tmin, tmax = -0.2, 0.8>>> reject = dict(grad=4000e-13, mag=4e-12, eog=180e-6)>>> baseline = None>>>>>> epochs = mne.Epochs(raw, events, events_id, tmin, tmax,

proj=True, picks=picks, baseline=baseline, preload=True, decim=2, reject=reject, add_eeg_ref=False)

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/05-make_epochs.html

…reject the ones with blinks

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/05-make_epochs.html


why did we drop epochs?Example of “drop log” for subject 1:


Use ICA on Epochs to remove ECG>>> ica_name = op.join(meg_dir, subject, 'run_%02d-ica.fif' % run)>>> ica = read_ica(ica_name)>>> n_max_ecg = 3 # use max 3 components>>> ecg_epochs = create_ecg_epochs(raw, tmin=-.5, tmax=.5)>>> ecg_inds, scores_ecg = ica.find_bads_ecg(ecg_epochs, method='ctps',

threshold=0.8)>>> ica.exclude += ecg_inds[:n_max_ecg]>>> ica.apply(epochs)

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_ica.html


Use ICA on Epochs to remove ECG



Fit and apply ICA on Epochs



From Epochs to Evoked

>>> evoked_faces = epochs['face'].average()>>> evoked_famous = epochs['face/famous'].average()>>> evoked_scrambled = epochs['scrambled'].average()>>> evoked_unfamiliar = epochs[‘face/unfamiliar'].average()>>> contrast = mne.combine_evoked([evoked_famous, >>> evoked_unfamiliar,>>> evoked_scrambled], >>> weights=[0.5, 0.5, -1.])>>> mne.evoked.write_evokeds('%s-ave.fif' % subject, [evoked_famous, evoked_scrambled, evoked_unfamiliar, contrast,

faces])

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/06-make_evoked.html

Index Epochs with tags

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/06-make_evoked.html

EEG Evoked for 3 first subjects

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_compare.html


EEG Evoked topographies (6 subjects)



MEG Evoked for 3 first subjects



Back to filtering…

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_fanning.html

>>> raw.filter(None, 40, fir_window='hann', phase='zero', h_trans_bandwidth=‘auto’, filter_length='auto')



>>> raw.filter(None, 40, fir_window='hann', phase='zero', h_trans_bandwidth=‘auto’, filter_length='auto')

Fanningproblem

Much lessfanning



>>> raw.filter(1, None, l_trans_bandwidth=0.5, fir_window='hann', phase='zero',

h_trans_bandwidth=‘auto’, filter_length='auto')

Covariance est. and whitening

Engemann, D.A., Gramfort, A., Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals., Neuroimage 2015

>>> cov = mne.compute_covariance(epochs, tmax=0, method='shrunk')

Covariance est. and whitening

>>> cov = mne.compute_covariance(epochs, tmax=0, method='shrunk')

whitened Global Field Power

whitened ERF

whitened ERP



Grand averaging

>>> for idx, evokeds in enumerate(all_evokeds):>>> # Combine subjects:>>> grand_avg[idx] = mne.combine_evoked(evokeds, 'equal')

>>> mapping = {'Famous': evokeds[0], 'Scrambled': evokeds[1], 'Unfamiliar': evokeds[2]}>>> mne.viz.plot_compare_evokeds(mapping, [idx])

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_group.html


Grand averaging

>>> for idx, evokeds in enumerate(all_evokeds):>>> # Combine subjects:>>> grand_avg[idx] = mne.combine_evoked(evokeds, 'equal')

>>> mapping = {'Famous': evokeds[0], 'Scrambled': evokeds[1], 'Unfamiliar': evokeds[2]}>>> mne.viz.plot_compare_evokeds(mapping, [idx])


From baseline to baseline !


MNE vs. Wakeman et al.



MNE vs. Wakeman et al.


Real?


Sensors stats (cluster level)

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_sensor_cluster_stats_eeg_channel.html

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_sensor_cluster_stats_eeg_channel.html

Sensors stats (cluster level)

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_sensor_spatio_temporal_cluster_stats.html

p = 0.01 (corrected)

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_sensor_spatio_temporal_cluster_stats.html

Decoding

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_time_decoding.html

>>> td = TimeDecoding(predict_mode='cross-validation', times=times, scorer='roc_auc')>>> td.fit(epochs, y)

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/08-run_time_decoding.html

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/plot_time_decoding.html

http://mne-tools.github.io/mne-biomag-group-demo/auto_scripts/08-run_time_decoding.html


• 3 Layers BEM forward modeling

• dSPM source localization on cortical surface

• loose orientation (0.2) & depth (0.8) & SNR = 3

• RMS across orientations

• morphing to fsaverage:subject fsaveragemorphing

… the MNE (default) way



Subject 2Contrast

localization (faces vs.

scrambled

Source space group analysis


16 subjectsaverage ofContrast (faces vs.

scrambled


MNE web report

http://perso.telecom-paristech.fr/~mjas/biomag_demo/report_sub002.html


MNE web report


Command line:$ mne report --path MEG/sub002 --info MEG/sub002/run_01_filt_sss-epo.fif --subject sub002 --subjects-dir subjects/ --verbose --jobs 6


Getting help http://martinos.org/mne/

http://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysisMailing list:


http://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

Getting inspired...http://martinos.org/mne/auto_examples/index.html


Sending feedbackhttps://github.com/mne-tools/mne-python


Some links• Documentation:

• http://martinos.org/mne/ (general doc)

• http://martinos.org/mne/stable/manual/index.html (manual)

• http://martinos.org/mne/stable/tutorials.html (tutorials with code)

• http://martinos.org/mne/auto_examples/index.html (python examples)

• Code:

• https://github.com/mne-tools/mne-python (mne-python code)

• https://github.com/mne-tools/mne-matlab (mne matlab toolbox)

• https://github.com/mne-tools/mne-scripts (mne shell scripts)


http://martinos.org/mne/stable/manual/index.html

http://martinos.org/mne/stable/tutorials.html

http://martinos.org/mne/auto_examples/index.html

https://github.com/mne-tools/mne-python

https://github.com/mne-tools/mne-matlab

https://github.com/mne-tools/mne-scripts

Access HCP-MEG data as MNE-Python data structures

https:/mne-tools.github.io/mne-hcp

http://mne-tools.github.io/mne-hcp

Seamlessly plug HCP MEG data into the Python ecosystem

…/mne-hcp/auto_tutorials/index.html

Poster

[Mo-P008]

Engemann et al.

Alexandre Gramforthttp://alexandre.gramfort.netContact:

GitHub : @agramfort Twitter : @agramfort

If you want a sticker come to the

MNE poster !

OpenScience!

http://alexandre.gramfort.net