Implementation of EEG pipeline

[sebastientourbier]

First thoughts

• Design/first sketch of the new EEG pipeline
• Function to load Cartool / EEGlab / Fieldtrip inverse solutions as input
• Implement the interface to compute single source dipoles per ROI based on SVD decomposition [Rubega et al. 2018] using pycartool
• Computes diverse common functional connectivity metrics (Imaginary coherence, ...) using MNE
• Implement the interface to compute time-varying functional connectivity based on the Granger causality framework and a modified version of the adaptive Kalman filter as proposed in [Pascucci et al. 2019]. This allows to recursively refine time-varying directed connectivity estimates with structural connectivity.

[Rubega et al. 2018] Rubega, M.; Carboni, M.; Seeber, M.; Vulliemoz, S.; and Michel, C. M. Estimating EEG source dipoles based on singular-value decomposition for connectivity analysis. 2018. paper

[Pascucci et al. 2019] Pascucci D.; Rubega M.; Hagmann P.; Plomp G. Adaptive filtering with anatomical priors: Integrating structural and effective connectivity. OHBM 19 abstract

What need to be answered

• How to visualize sources estimated by Cartool in MNE (translation in log file enough?)
• Check if MNE provides already function to load EEGlab, FieldTrip
  Yes (TO DO: list the functions).
• How Cartool / EGGlab / Fieldtrip / Brainstorm / MNE are saving outputs (BIDS or in-house organization). This will indicate us how to grab the data (cleaned EEG data and LeadField matrix) in the most automatical way.
• The determination of a common source space (Freesurfer space might be prefered as we already provide the annotation files for the Lausanne2008 and Lausanne2018 parcellations)
• Difference between SVD method in [Rubega et al. 2018] and the SVD method implemented in MNE (mode: pca_flip)
• How to adopt BIDS EEG (see spec paper)

To be prepared and implemented

Note: check in Nipype which MNE interfaces are already implemented which could then be used as they are or as a template for the developement of the MNE-related interfaces (Tasks 5. and 6.)

Achievement

• Task 1
• Task 2
• Task 3
• Task 4
• Task 5
• Task 6
• Task 7
• Task 8
• Task 9

Task 1

Preparation of a sample dataset for development (See #6 (comment) for more details)

Task 2

Start the implementation of a simple Nipype workflow that can grab and convert to MNE's internal format the cleaned EEG data and the Lead Field matrix generated by:

• Cartool
• EEGlab
• Fieldtrip
• MNE
• Brainstorm

Task 3

Implement for each toolbox an interface after the data grabber that converts and stores in MNE format the generated outputs (cleaned EEG data and lead field matrix)

Inputs

• Cleaned EEG in toolbox format
• Lead field matrix in toolbox format

Outputs

• Cleaned EEG in MNE format
• Lead field matrix in MNE format

Task 4

Implement an interface that creates the inverse solutions and transforms the source coordinates into a common source space

Inputs

• Cleaned EEG in MNE format (channels x time)
• Lead field matrix in MNE format (#sources x channels)
• Transformation from toolbox source coordinates to common space

To be performed

• Computes inverse solution (sources):
  Source = Lead field matrix x cleaned EEG
• Transform the source coordinates (sources)

Outputs

• Source time-series
• Source coordinates

Task 5

Check if output sources from the different toolbox are in proper common space

Task 6

Implementation of an interface for estimating ROI dipoles based on Maria's SVD and/or MNE SVD methods.

Inputs

• Inverse solution in MNE format
• Parcellation annotation files or parcellation scheme

Outputs

• ROI dipoles in MNE format

To be performed

• Use MNE function or implementation of Maria's method to create ROI dipoles in MNE format

Task 7

Implementation of an interface for estimating multiple dynamic functional connectome maps

Inputs

• ROI dipoles in MNE format

Outputs

• Dynamic FC maps in .gpickle format
• Dynamic FC maps in .mat format

To be performed

• Computes multiple dynamic functional connectome maps using measure implemented in MNE and pycartool

Task 8 (Large effort)

Implementation of EEGPipeline:

• Organisation of pipeline and stages (for configuration and ouput inspection with bidsappmanager)

• Implement the stages (config parameters, create_worflow, inpsect_outputs) and modify the existing cmp/pipelines/EEG.py to implement the complete workflow

• Implement the GUI components (EEGPipelineUI and all the related stages) for integration in the cmpbidsappmanager

• Review bidsapp parser and run.py to take as input the eeg config file

Task 9

Documentation:

• Add functionality in the index.html page
• Update the bidsappmanager.html with EEG in 1) pipeline configuration, 2) run the bidsapp and 3) check stage outputs,
• Update outputs.html page with BIDS-EEG derivatives data produced by CMP

[sebastientourbier]

Code providing cartool I/O and functions in python:

• https://github.com/Functional-Brain-Mapping-Laboratory/PyCartool/tree/master/pycartool
• https://pycartool.readthedocs.io/en/latest/?badge=latest

[sebastientourbier]

Investigate head model with MNE

[sebastientourbier]

What need to be answered

* [ ]  Check if MNE provides already function to load EEGlab, FieldTrip
  Yes (TO DO: list the functions).

* [ ]   How Cartool / EGGlab / Fieldtrip / Brainstorm / MNE are saving outputs (BIDS or in-house organization). This will indicate us how to grab the data (cleaned EEG data and LeadField matrix) in the most automatical way.

* [ ]   The determination of a common source space (Freesurfer space might be prefered as we already provide the annotation files for the Lausanne2008 and Lausanne2018 parcellations)

* [ ]  Difference between SVD method in [Rubega et al. 2018] and the SVD method implemented in MNE (mode: `pca_flip`)

* [ ]  How to adopt BIDS EEG ([see spec paper](https://www.nature.com/articles/s41597-019-0104-8.pdf))

To be implemented

1. A simple workflow that can grab and convert the MNE internal format the cleaned EEG data and the Lead Field matrix generated by:


* [ ]   Cartool

* [ ]  EEGlab

* [ ]  Fieldtrip

* [ ]   MNE

* [ ]   Brainstorm


1. Implement for each toolbox an interface after the data grabber that converts and stores in MNE format the generated outputs (cleaned EEG data and lead field matrix)

Inputs

* Cleaned EEG in toolbox format

* Lead field matrix in toolbox format

Outputs

* Cleaned EEG in MNE format

* Lead field matrix in MNE format


1. Implement an interface that creates the inverse solutions and transforms the source coordinates into a common source space

Inputs

* Cleaned EEG in MNE format (channels x time)

* Lead field matrix in MNE format (#sources x channels)

* Transformation from toolbox source coordinates to common space

To be performed

* Computes inverse solution (sources):
  Source = Lead field matrix x cleaned EEG

* Transform the source coordinates (sources)

Outputs

* Source time-series

* Source coordinates


1. Check if output sources from the different toolbox are in proper common space


* [ ]   Cartool

* [ ]  EEGlab

* [ ]  Fieldtrip

* [ ]   MNE

* [ ]   Brainstorm


1. Implementation of an interface for estimating ROI dipoles based on Maria's SVD and/or MNE SVD methods.

Inputs

* Inverse solution in MNE format

* Parcellation annotation files or parcellation scheme

Outputs

* ROI dipoles in MNE format

To be performed

* Use MNE function or implementation of Maria's method to create ROI dipoles in MNE format


1. Implementation of an interface for estimating multiple dynamic functional connectome maps

Inputs

* ROI dipoles in MNE format

Outputs

* Dynamic FC maps in .gpickle format

* Dynamic FC maps in .mat format

To be performed

* Computes multiple dynamic functional connectome maps using measure implemented in MNE and pycartool

Based on our discussion with Joan yesterday

[sebastientourbier]

Preparation of a sample dataset for development

Tasks involved

• Decision and creation of a common dataset (source data: T1w, EEG, (DWI and fMRI would be great also))

• Estimation of inversion solutions for:
  • Cartool (proprietary)
  • EEGlab (open / matlab)
  • Fieldtrip (open / matlab)
  • Brainstorm (open / matlab)
  • MNE (open / python)
• Organization of the different outputs in the dataset:
  • Cartool (proprietary)
  • EEGlab (open / matlab)
  • Fieldtrip (open / matlab)
  • Brainstorm (open / matlab)
  • MNE (open / python)

Example of structure

   ds-example/
    
    README
    CHANGES
    participants.tsv
    dataset_description.json
    
     sub-<subject_label>/
        anat/
            sub-<subject_label>_T1w.nii.gz
            sub-<subject_label>_T1w.json
        ...
        eeg/
           [To be checked in BIDS-EEG]
    
    ...

    derivatives/
        cartool/
            sub-<subject_label>/
                eeg/
                    [Outputs from cartool]

        eeglab/
            sub-<subject_label>/
                eeg/
                    [Outputs from eeglab]

        fieldtrip/
            sub-<subject_label>/
                eeg/
                    [Outputs from fieldtrip]

        mne/
            sub-<subject_label>/
                eeg/
                    [Outputs from mne]

    ...

Resources

• BIDS-EEG: https://www.nature.com/articles/s41597-019-0104-8
• BIDS-EEG derivatives: https://github.com/bids-standard/bids-specification/issues/317

[sebastientourbier]

This is link to ohbm brainhack project "Integrate EEG inside CMP3" (ohbm/hackathon2020#214)