-
Notifications
You must be signed in to change notification settings - Fork 0
GSOC Ideas
Important: Expectations for prospective students
Possible Mentor: Alex Gramfort, Denis Engemann, Eric Larson, Martin Luessi
Possible Candidate: Roman Goj (University of Stirling)
Goal: MNE-python implement time domain beamformers (LCMV). The goal is to extend the beamformer module to support frequency domain implementations
References
Possible Mentor: Alex Gramfort, Martin Luessi, Christopher Dinh, Lauri Parkkonen
Possible Candidate: Mainak Jas (Aalto University, Helsinki)
Goal : The objective of this project is to setup a real-time Magnetoencephalography (rtMEG) processing pipeline able to do standard preprocessing up to supervised learning (classification) using the scikit-learn project (http://scikit-learn.org).
References
- http://prezi.com/p8vsx2glmaab/meg-real-time-acquisition-processing/
- http://www.hindawi.com/journals/cin/2011/327953/cta/
Possible Mentor: Alex Gramfort
Possible Candidate: ???
Goal: FreeSurfer provides segmentations of deep brain sructures (aseg volume segmentation). The goal of this project would be to have source spaces restricted to these structures and to combine them with surface based source spaces that MNE commonly uses.
Possible Mentor: Denis Engemann, Alex Gramfort
Possible Candidate: ?
Goal : Implement fMRI style statistical modelling of mixed effects including crossed random effects. A GLM(E) for MEG would allow for creating flexible parametric mass-univariate models of observed MEG data. Besides a standard gaussian link-function, template functions for evoked and induced responses might be incorporated in such a model. On the other hand side, modelling crossed random effects allows to test interaction terms in commonly used so called 'repeated measures'-designs (e.g., randomized block designs). The new functionality, hence, could be used to generate first-level stats and / or to formulate sophisticated contrasts at the group level. For group analyses, it could be passed as stat_fun to our existing clustering permutation test. Finally, this also might implicate developing better tools for continuous single-trial regression and to complement existing t-tests for factorial designs.
Some projects that would provide building blocks are:
- NiPy, which recently added a mixed effects implementation: http://goo.gl/54Mb2
- Patsy (formula parsing and creation of design matrices): http://patsy.readthedocs.org/en/latest/
- Statsmodels (general resource): http://statsmodels.sourceforge.net/devel/index.html
- lme4 (R), (general reference): http://cran.r-project.org/web/packages/lme4/index.html
Possible Mentor: Alex Gramfort, Eric Larson
Possible Candidate: Junaid Naseer
Goal : MNE-python does not provide yet routines for forward modeling. The ambition of this project is to port/bind to Python existing C code for forward modeling (Allow setup source space and compute gain matrix).
Possible Mentor: Alex Gramfort, Eric Larson
Goal: With all the demos of python web based viz (Flask with d3.js, IPython notebook hooks, galry, etc.) it could be doable to have convenient viz tools runnning in a browser for easy remote access.
See :
- https://github.com/ContinuumIO/Bokeh
- https://github.com/rossant/galry
- http://graphics.cs.brown.edu/research/sciviz/newbraininteraction/
- https://github.com/vegardlarsen/MRI-viewer
Possible Mentor: Alex Gramfort, Denis Engemann
Possible Candidate: ???
Goal: The ambition is to have simple tools to review the quality of a data analysis and report the output as web pages. Something like mne_report.py --subject sample --raw sample_audvis_raw.fif --bem ... --ave sample_audvis-ave.fif Among other steps the report would include artifact rejection quality using SSP or ICA, checking assumptions for ERF analysis, sensitivity maps, etc. See https://github.com/neurospin/pypreprocess/tree/master/reporting for inspiration.