What's next
A roadmap to the future developments of Brainstorm.
Contents
Current topics
Source modeling
- Implementation of a new unified minimum norm/beamformer framework (almost finished)
Documentation
Cross-validation with MNE and FieldTrip (work in progress):
http://martinos.org/mne/dev/auto_tutorials/plot_brainstorm_auditory.html
Functional connectivity
- Significance thresholding of the connectivity matrices (not started)
Computation
- Removing the dependence to the Java interface to run in headless mode (not started)
Call MNE-Python and FieldTrip functions from Brainstorm GUI (work in progress)
Recordings
RAW file viewer:
- Downsample before filtering? (attention to the filter design)
- Add parameter to make the visual downsampling more or less aggressive
- Pre-load next page of recordings
- Keep the filter specifications in memory instead of recomputing for every page
- Add field "comment" to markers: For clinicians to add notes (Marcel)
- Events: Change the category of a selected event easily, instead of deleting/marking new
Events: Advanced process for recombining.
Example: http://www.erpinfo.org/erplab/erplab-documentation/manual/Binlister.html
- Bad trials: When changing the status of bad to good: remove the bad segments as well, otherwise it is not processed by processes like the PSD.
- Review clinical recordings: Reduce the dimensionality of the data with a simple inverse problem, similar to what we do for the magnetic extrapolation ("Regional sources" in BESA, cf S Rampp)
- MEG/EEG registration: Apply the same transformation to multiple runs
- 2DLayout:
- Overlay multiple conditions
- Does not work when DC offset is not removed
- Add a proper amplitude scale that gets updated when shift+scroll, to compare figures
- Update the vertical lines when changing the gain (shift+scroll)
Interface
Add a warning when computing a forward model with > 100000 sources (check selection)
- Start Brainstorm without Java (-nodesktop)
- Option to increase the size of all the elements of the interface (for high res laptop screens)
- Snapshot: Save as image / all figures (similar to Movie/all figure)
Generalize the use of the units (field .DisplayUnits): Rewrite processes to save the units correctly
- Colormaps:
- Allow brightness/contrast manipulations on the custom colormaps
- Create a colormap similar to MNE, where extrema are bright
- Global colormap max: Should get the maximum across all the open files
- Open new figures as tab (docked in the Figures window)
- Copy figures to clipboard (with the screencapture function)
- Removing all the CTRL and SHIFT in the keyboard shortcuts
- Display warning before opening files that are too big
- Smooth display from figure_image (ERPimage, raster plot...)
Contact sheets & movies: use average of time windows instead of single instants, for each picture.
- Contact sheets: Allow explicit list of times in input (+ display as in MNE-Python with TS)
- Display CTF coils: Show discs instead of squares
- Use boundary() instead of conhull() in all the display functions (ie. 2DDisc)
Connectivity
- Thresholding and stat tests the connectivity matrices
- Connectivity on unconstrained sources: "Default signal extraction for volume grids should be the time series of the first principal component of the triplet signals after each has been zero-meaned" (SB)
- Display of connectivity graphs:
- Display as straight lines
- Recode 2D graphs
- 3D display with anatomical constrains
- Use new band-pass filters in bst_connectivity ('bst-hfilter' instead of 'bst-fft-fir')
Review by Jan-Mathijs: http://journal.frontiersin.org/article/10.3389/fnsys.2015.00175/full
- Connectivity based on band limited power (Sylvain):
- Compute Hilbert/Bandpass + correlation of the envelopes
- Bandpass envelopes before computing correlations?
- Compute Hilbert(sensors) and then project to source space?
- Graph view:
- Does not display negative values correctly (correlation or difference of coherence)
- Re-write using pure Matlab code and smoothed graphics
- Fixed scales for intensity sliders
- Text bigger
- Too much data in appdata
- Fixed scales for intensity sliders
- Add "=" shortcut for having graphs with similar configurations
- Disable zoom in one region (serious bugs)
- NxN on sensors: does not place the sensors correctly in space
- Coherence:
Average cross-spectra instead of concatenating epochs (to avoid discontinuities)
Explore inter-trial approaches (Esther refers to chronux toolbox)
- Granger:
- Crashes sometimes: improve stability
- Check for minimum time window (Esther: min around 500-1000 data points)
- Re-write and optimize code
- Add progress bar
- PLV:
- Add p-values
- Remove evoked
- Optimize code
- Add time integration
- Unconstrained sources
- Add warning when running of short windows (because of filters)
- PAC:
- Add input TF , to disconnect TF decomposition and PAC computation (Peter)
- Refine frequency vector of low frequencies
- How many central frequencies to use in bst_pac?
- Change filters: no chirplet functions
- bst_freqfilter: Use nfcomponents like in bst_pac
- Esther recommended a larger frequency binning of the PAC estimation
- PAC maps: Display all sensors at once (like TF and DynamicPAC)
- Hui-Ling's PAC:
- Time-resolved correlation/coherence: Display as time bands
- Other metrics:
- Coherence by bands: bst_coherence_band_welch.m
- Granger by bands: bst_granger_band.m
- Inter-trial coherence
Tutorial coherence [1xN] : Reproduce FieldTrip results?
Connect NxN: Display as time series > Display warning before trying to open too many signals
Processes
- Use CUDA for speeding up some operations (filtering, wavelets, etc)
- Allow processes in Python and Java
- Add MNE-Python functions:
- scikit-learn classifiers: S Marti / G Dehaene
- Implement data exchange with MNE-Python: write FIF files from Brainstorm and/or pass python objects in memory instead of FIF files
- SSS/tSSS cleaning
- Reproduce other tutorials / examples
- Change the graphic renderer from Matlab
Add FieldTrip functions:
- ft_sourceanalysis:
- Check noise covariance
- Check all the options of all the methods
- Single trial reconstructions + noise covariance?
Filters?? http://www.fieldtriptoolbox.org/example/common_filters_in_beamforming
Beamformers: Save ftSource.avg.mom
http://www.fieldtriptoolbox.org/workshop/meg-uk-2015/fieldtrip-beamformer-demohttp://www.fieldtriptoolbox.org/tutorial/beamformingextended
- Baseline? Two inputs?
ft_prepare_sourcemodel: Compute MNI transformation (linear and non-linear) => Peter
- Freqanalysis: ITC
- ft_read_atlas('TTatlas+tlrc.BRICK');
ft_volumereslice: http://www.fieldtriptoolbox.org/faq/how_change_mri_orientation_size_fov
- ft_freqanalysis
- ft_combineplanar
Allow FieldTrip functions in compiled version
- ft_sourceanalysis:
- Decoding/Classifiers: Faster algorithms (MNE-Python?)
- Pipeline editor:
- Bug: After "convert to continuous", the time of the following processes should change
- Add loops over subjects/conditions/trial groups
ITC: Inter-trial coherence (see MNE reports for group tutorial)
http://www.sciencedirect.com/science/article/pii/S1053811916304232- ICA:
- Why doesn't the ICA process converge when using 25 components in the EEG tutorial?
- Add an option to resample the signals before computing the ICA decomposition
- Add a stand-alone tutorial
- Exploration: Add window with spectral decomposition (useful for muscle artifacts)
- Export IC time series (and then compute their spectrum): solves the problem above
Comparison JADE/Infomax:
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0030135- Use faster methods (MNE-Python?)
- Add methods: SOBI, Fastica, AMICA/CUDICA (recommended by S Makeig)
Dimension reduction with PCA adds artifacts: Not done by default in EEGLAB
Contact: Stephen Shall Jones ( shall-jones@infoscience.otago.ac.nz )
Student Carl Leichter detailed this in his thesis- S Makeig: Use ICA to select the IC of interest instead of only removing artifacts
- Display of spectrum for components (PSD/FFT)
- Use FastICA (algo crashing)
- Add components preselection: Correlation with EOG/ECG
- Import ICA matrices available in EEGLAB .set files
EEGLAB recommends ICA + trial rejection + ICA again: Impossible right now with Brainstorm
(http://sccn.ucsd.edu/wiki/Chapter_09:_Decomposing_Data_Using_ICA)ICA+machine learning: https://www.ncbi.nlm.nih.gov/pubmed/28497769
Automated artifact rejection: https://arxiv.org/abs/1612.08194
- Save IC time series in database
- SSP:
Display warning if changing the ChannelFlag while there is a Projector applied
Show where the attenuation is projected:
(sum(IK,2)-sum(SSP(k,:)*IK,2)./sum(IK,2)
- Pipeline editor:
- When computing sources from the pipeline editor: doesn't reselect the options if you click twice on "edit" (works for minnorm, but not for lcmv)
- When computing time-frequency/hilbert/psd: Find a way not to force the user to click on Edit
- Bandpass:
- Use new filters in all the functions using a bandpass ('bst-hfilter' instead of 'bst-fft-fir'): process_evt_detect_threshold
- Weird bug: Filter(import) != Import(Filter) in the HCP tutorial... to investigate
- Bandpass
- Spectral flattening (John):
- ARIMA(5,0,1): Apply on the signal before any frequency/connectivity/PAC analysis
- PSD:
- Rewrite to have the same input as coherence (frequency resolution instead of window length)
- Use the progress bar
Allow display of Avg+StdErr
Remove line noise: http://www.nitrc.org/projects/cleanline
- Use Matlab Coder to optimize some processes: Wavelets, bandpass filter, sinusoid removal
- Reports: Click on link reopens exactly the figure
- Time-frequency:
- Optimization: bst_timefreq (around l.136), remove evoked in source space: Average should be computed in sensor space instead of source space (requested by Dimitrios)
Short-time Fourier transform: http://www.mikexcohen.com/lectures.html
Matching pursuit: http://m.jneurosci.org/content/36/12/3399.abstract?etoc
- Bug: Display logs as negative
- Bug: 3D figures: Colormaps with "log" option doesn't work
- Bug: Difference of power displayed in log: problems (Soheila)
- 2D Layout in spectrum
- Make much faster and more memory efficient (C functions coded by Matti ?)
- TF scouts: should display average of TF maps
- Impossible to keep complex values for unconstrained sources
- Pad short epochs with zero values for getting lower frequencies
- Hilbert with time bands very slow on very long files (eg. 3600s at 1000Hz) because the time vector is still full (10^7 values): save compressed time vector instead.
- Extend clusters tab to display of TF to overlay TF signals (Svet)
- When normalizing with baseline: Propagate with the edge effects marked in TFmask
- Allow running TF on montages
- Review continuous files in time-frequency space (for epilepsy)
- Bug when computing TF on constrained and unconstrained scouts at the same time (in mixed head models for instance): uses only the constrained information and doesn't sum the 3 orientations for the unconstrained regions.
- Artifact detection:
- Artifact rejection like SPM: if bad in 20%, bad everywhere
- Test difference between adjacent samples
- Events detection: Add option "std" vs "amplitude"
- Co-registration of MEG runs:
- SSP: Group projectors coming from different files
- Finish validation of the method
- Apply to continuous recordings for correcting head movements
- Other processes:
- Moving average
- Max
- Median
- Simulation:
Fix units in simulation processes => no *1e-9 in "simulate recordings"
- Use "add noise" process from Hui-Ling (in Work/Dev/Divers)
- Use field process field "Group" to separate Input/Processing/Output options
- Use new Matlab functions: movmean, movsum, movmedian, movmax, movmin, movvar, movstd
- Process ft_prepare_heamodel: Add support from BEM surfaces from the Brainstorm database
Database
- Add buttons to sort files: by name, by comment, by date
- MEG protocols: More flexible organization of the database; sub-conditions to allow different runs X different conditions.
- Matrix files: Allow to be dependent from other files
- Add notes in the folders (text files, visible as nodes in the tree)
- Screen captures: save straight to the database
- Rename multiple files
- Default headmodel lost when reloaded: Keep selection on the hard drive (in brainstormstudy.mat)
- protocol.mat can be too big: do not store the results links in it (and recreate when loading)
Distributed computing
Options from FieldTrip:
Loose collection of computers: https://github.com/fieldtrip/fieldtrip/tree/master/peer
Alternative, with less limitations: http://research.cs.wisc.edu/htcondor/
Single multicore machine: https://github.com/fieldtrip/fieldtrip/tree/master/engine
Batch system: https://github.com/fieldtrip/fieldtrip/tree/master/qsub
Documentation: http://fieldtrip.fcdonders.nl/faq#distributed_computing_with_fieldtrip_and_matlab
Various initiatives: http://samirdas.github.io/Data_sharing.html#/
Source modeling
Menu head model > Copy to other conditions/subjects (check if applicable first)
Menu Sources > Maximum value: Doesn't work with volume or mixed head models
Mixed head models:
- Set loose parameter from the interface
- Volume grid:
- Optimize: 3D display (better than 9x9 cubes)
- Optimize: vol_dilate (with 26 neighbors)
Menu Sources > Simulate recordings:
- Do not close the 3D figures after generating a new file
- Add a process equivalent to this menu
- Panel Get coordinates: Add button "find maximum"
- Dipoles:
- Project individual dipoles files on a template
- panel_dipoles: Doesn't work with multiple figures
- BEM single sphere: Get implementation from MNE
- Visualize Beamformer results:
- Read CTF SAM .svl
- Display as layers in the MRI viewer
- Unconstrained sources:
- Stat and connectivity: what to do? (re-send email John+Sylvain)
- Overlapping spheres: improve the estimation of the spheres for the frontal lobes
- Magnetic extrapolation: Do the same thing with EEG
- Noise covariance matrix:
- Display with figure_image()
- Storage of multiple noise covariance matrices (just like the head models)
- Always save as full, then at inversion time, we can decide between full, heteroskedastic (diagonal) or homoskedastic (i.i.d, scalar)
Problem of having inividual trials + averages in the condition => Display warning or not?
- Save nAvg in noisecov file, to make it easier to scale to other recordings
Sources on surface: Display peak regions over time (time = color) => A.Gramfort
Calculate ImagingKernel * Gain for a scout
- Time-frequency beamformers:
- Band-pass everything in different frequency bands + Source estimation + TF
- Ask data to Sarang where he sees effects that cannot be extracted with MN followed by TF
- Process "Extract scouts time series": Add PCA option (replace isnorm with choice PCA/Norm)
- BEM: Fix unstable results when one vertex is too close from the layers (5mm ?)
- Hui-Ling beamformers:
- More explanations about what is in NAI and Spatial filters
Explain that is this is better to study effects extended in time (Ntime > Nsensors)
- Group LCMV+MCB
- Condition LEFT median nerve: very bad results
- Keep options for inverse computation
- OpenMEEG: Post example datasets for the remaining issues:
- Example protocol ECOG: doesn't work
- Add eyes models to attract eye activity
Anatomy
- Registration:
Use the same registration for multiple recording sessions that have already re-registered previously (eg. with MaxFilter)
- Check templates: MNI transformation and volume atlases for ICBM152 vs Colin27 (loading the AAL atlas as surface or voume scouts do not align well on the ICBM152)
- Compute non-linear MNI registration instead of linear
- Select and remove bad digitized head points before automatic coregistration
Load the MNE -transf.fif: http://neuroimage.usc.edu/forums/showthread.php?2830
- MRI Viewer:
- Pan in zoomed view (shift + click + move?)
- Zoom in/out with mouse (shift + scroll?)
- Ruler tool to measure distances
- Add keyboard shortcuts to scroll in the three orientations (same in MRI 3D)
- Display scouts as additional volumes
- Render surface envelope in the MRI as a thin line instead of the full interpolation matrix
- Edit fiducials: Replace 6 text boxes with 1 for easy copy-paste (see fiducials.m)
- Optimize computation interpolation MRI-surface (tess_tri_interp)
- Add colormap for atlases / integers (FSL "random-rainbow" or MRIcron "rainramp")
BrainVISA: Add support for MarsAtlas (Guillaume A)
BrainSuite:
Use BrainSuite inner skull for surface generation
- Use same colors for left and right for anatomical atlases
- Scouts:
- Display edges in the middle of the faces instead of the vertices
- Display scouts in a tree: hemisphere, region, subregion
- Sort scouts by region in process options
- Downsample to atlas: allow on timefreq/connect files
- Parcellating volume grids: scikit-learn.cluster.Ward
- Major bug when importing surfaces for an MRI that was re-oriented manually
- Smooth surface: Fix little spikes due to irregularities in the mesh
- Bug: Hide scouts in the preview of the grid for volume head models
ECOG/SEEG
Contact positions: Import / set / detect
- Automatic segmentation of CT scans: See project of Rodrigo Paz (Marseille/Timone)
Import from FreeSurfer/FreeView: Problem of difference between RAS and TkRegRAS: http://neuroimage.usc.edu/forums/showthread.php?1958-SEEG-electrodes-and-subject-s-anatomy-are-not-alligned
- Add history: Save modifications and transformations applied to the channel files (Marcel)
- ECOG: How to handle cases where not all the grid contacts are in the channel file? (Marcel)
- Project contact positions across subjects or templates (Marcel)
- Display:
- Display SEEG+ECOG contacts at the same time
- Better way to represent ECOG strips that are not in contact with the skull (new type?)
- SEEG length: count from the most superficial contact
- ECOG: Project on cortex instead of inner skull, just for display (Marcel)
- Color each strip/grid differently
- Bad channels: Contacts greyed out instead of ignored (Marcel)
- ECOG: Topography display without the 3D coordinates (Marcel)
- 2DLayout: Should not try to use the actual position of the sensors, just the grids/strips.
- Display time in H:M:S
Alternatives to OpenMEEG: SimBio/FieldTrip? Matti Stenroos? NFT/NIST?
- Prepare tutorial with public dataset (Marcel)
Statistics
- ANOVA:
- Use LENA functions(?)
- Output = 1 file per effect, all grouped in a node "ANOVA"
- Display several ANOVA maps (from several files) on one single figure, using a "graphic accumulator", towards which one can send any type of graphic object
- New process to test for Gaussianity using swtest
- PLS: Partial Least Squares
- Simulate recordings with specific properties, for stat validation
Quality control before statistics, on condition averages across subjects:
mean(baseline)/std(baseline): shows bad subject quickly.Use SurfStat: Impements interesting things, like an analytical cluster-based p-value correction (Random-field theory which is used in SPM) - Peter
Input / output
XDF import: Use the EEGLAB plugin, contact Martin Bleichner (Oldenburg)
- DICOM converter:
- Add dcm2nii (MRICron)
- Add MRIConvert
FieldTrip: Import/Export time-frequency:
- 4D file format:
- Use reader from MNE-Python: mne.io.read_raw_kit (doesn't require Yokogawa slow library)
- Reference gradiometers: Keep the orientation of the first or second coil?
- Reference gradiometers: Add the sensor definition from coil_def.dat
- Validate with phantom recordings that noise compensation is properly taken into account
- References at too far from the head sensors in Marseille 4D system
- The noise compensation is considered to be always applied on the recordings, not sure this assumption is always correct
- 4D phantom tutorial (JM Badier?)
EEG File formats:
- Nicolet EEG: Get from from MNE
EEG CeeGraph
- EGI: Finish support for epoched files (formats 3,5,7)
- BCI2000 Input (via EEGLAB plugin)
- EEGLAB import:
- Support for binary AND epoched files (now it's one or the other)
- Allow epoched files with recordings saved in external files
- BST-BIN: Add compressionto .bst
Persyst .lay: https://github.com/ieeg-portal/Persyst-Reader
Nervus .eeg: https://github.com/ieeg-portal/Nervus-Reader
Biopac .acq: https://github.com/ieeg-portal/Biopac-Reader
- Review raw on all the file formats (ASCII EEG and Cartool missing)
- gTec EEG recordings: Read directly from the HDF5 files instead of the Matlab exports.
- Use new Matlab functions readtable/writetable (2014a): for Excel and text files
- SPM .mat/.dat: Fix the import of the EEG/SEEG coordinates
Distribution & documentation
- Add tags to the forum posts for easier listing by topic
- Tutorial OMEGA/BIDS:
- Add review of literature for the resting state MEG
- Download example datasets directly from the OMEGA repository
- New tutorials:
Visual group Henson/Wakeman tutorial: Use the BIDS version
https://github.com/INCF/BIDS-examples/tree/bep008_meg/ds000117Other public datasets: https://github.com/INCF/BIDS-examples/tree/bep008_meg/
- Rat PAC + high gamma (Soheila)
- EEG/research
FieldTrip ECOG tutorial: http://www.fieldtriptoolbox.org/tutorial/human_ecog
FieldTrip cortico-muscular coherence tutorial: http://www.fieldtriptoolbox.org/tutorial/coherence
Reproduce tutorials from MNE-Python: https://martinos.org/mne/stable/tutorials.html
Cam-CAN database: https://camcan-archive.mrc-cbu.cam.ac.uk/dataaccess/<<BR>>(download new datasets, including maxfiltered files and manual fiducial placements)
- MEG steady-state / high-gamma visual / frequency tagging
- BIDS-EEG example datasets
- Intra-cranial recordings (Marcel, Jeremy?)
- Stand-alone ICA tutorial
- Co-register MEG runs (Beth)
- Move all the files to download on the cloud for faster download everywhere in the world
- Provide secure way of sending password over HTTPS for:
- Account creation
- Forum exchanges
org.brainstorm.dialog.CloneControl
Workflows FieldTrip: http://www.fieldtriptoolbox.org/faq/what_types_of_datasets_and_their_respective_analyses_are_used_on_fieldtrip
Reference tutorials on Google scholar + ResearchGate
Count GitHub clones in the the download stats
- Google Analytics: Create template and update the section of the Community page
- Missing in page "Cite Brainstorm": Add all the methods used in the software
- Deface the MRIs of all the tutorials
- Compiled R2016b: Color picker doesn't work (for changing surface color for instance)
Current bugs
- Screen capture:
- Bug on Win8/Win10: doesn't capture the correct part of the screen
- Window managers with fading effect: captures the top window
- Image viewer:
- Difficult to get to 100%
- Buggy on some systems
- 2DLayout:
- (TF) Units are weird with % values
- (TF) Difficult to navigate in frequencies: Scaling+changing frequency resets the scaling
- Progress bar:
- Doesn't close properly on some Linux systems
- Focus requests change workspace when processing constantly (Linux systems)
- MacOS bugs:
- Buttons {Yes,No,Cancel} listed backwards
- Record tab: Text of epoch number is too big
- Colormap menus: Do not work well on compiled MacOSX 10.9.5 and 10.10
- Matlab bugs:
- Interface looks small on screens with very high resolutions: Reduce the resolution
- Event markers are not visible anymore with the sequence: Open MEG, open EOG, close MEG
- in_bst_data_multi: If trials have different sizes, output is random (the one of the first file)
- Edit scout in MRI: small modifications cause huge increase of the scout size
- Canolty maps computation: Fix progress bar
Geeky programming details
- bst_bsxfun: After 2016b, we can use directly the scalar operators (./ .* ...) instead of bsxfun. Update bst_bsxfun to skip the use of bsxfun when possible.
Interface scaling: Rewrite class IconLoader to scale only once the icons at startup instead of at each request of an icon (might improve the speed of the rendering of the tree)
- Hide Java panels instead of deleting them
- Processes with "radio" and "radio_line" options: Replace with "radio_label" and "radio_linelabel"
- Interpolations: Use scatteredInterpolant, griddedInterpolant, triangulation.nearestNeighbor (2014b)
- bst_warp and channel_project: Use tess_parametrize_new instead of tess_parametrize
- Shared kernels: "get bad channels" operation in a different way (reading all the files is too slow)
- Optimize bst_get:
- Now study and subject have necessarily the same folder name
- Replace big switch with separate functions
- Progress bar:
- Add different levels (to handle sub-processes)
- Make work correctly with RAW on resting tutorial
- Uniformize calls in bst_process/Run
- Add a "Cancel" button
- Fix all the 'todo' blocks in the code
- Error message: Add a link to report directly the bug on the forum