Multivariate Pattern Analysis (MVPA) for fMRI
In recent years multivariate methods of data analysis have received increasing interest, in particular for the analysis of structural and functional magnetic resonance imaging (fMRI) data. Typical MRI analysis methods observe each voxel separately and describe general differences between conditions. In contrast, multivariate methods – also known as multivariate pattern analysis (MVPA) or decoding – allow access to information which can only be attained by taking into account multiple voxels at the same time. This makes it possible to ask questions about the specificity of representations in the human brain. In that way, new questions about neuronal representations in the brain can be addressed, for example which word a person was reading or which object is being held in memory.
This two-day course serves as an introduction into the multivariate analysis of fMRI data. Apart from a general introduction we will provide a comprehensive introduction to supervised classification (including support vector machines and cross-validation), advantages and disadvantages of whole brain vs. ROI vs. searchlight decoding, how to design MVPA studies, preprocessing, feature and parameter selection, correct statistics (and overlooked facts about null distributions in cross-validation), basic representational similarity analysis, and caveats that are important to bear in mind (such as why there can be below-chance results). The practical parts are carried out with the SPM-compatible Matlab toolbox "The Decoding Toolbox (TDT)" that we developed and which is a highly versatile and easy to use software package.
Instructors
Kai Görgen (AG Haynes, BCCN Berlin)
Number of participants
max. 24
Duration
04. Sep - 05. Sep daily 09:00-17:00 Uhr
Prerequisits
Basic knowledge on fMRI data analysis(no newcomers), basic programming in matlab (for-loops, matrix manipulation, functions, etc.); see the matlab tutorial at the end of this page
Helpful but not mandatory
If possible, bring your laptop (Linux, Windows oder Mac), with Matlab, SPM8 or 12, TDT and MRICroN (download links below) installed. We cannot guarantee VPN-access, but will do our best to provide network access if you do not have a matlab standalone license.
Suggested reading
Hebart, Görgen, and Haynes: "The Decoding Toolbox (TDT): A versatile software package for multivariate analyses of functional imaging data". (2015) Front. Neuroinform. 8:88, doi: 10.3389/fninf.2014.00088
Haynes and Rees: "Decoding mental states from brain activity in humans". (2006) Nature Reviews Neuroscience 7, 523-534.
Pereira, Mitchell, and Botvinick: "Machine learning classifiers and fMRI: a tutorial overview". (2009); Neuroimage 45, S199-S209.
Weblinks
Matlab-Tutorial (Punkte 1 bis 9)
Registration
We kindly ask all German speakers to register using this
registration form
by email or Fax +49 (0) 40 7410 59955 .
English speakers, please send us a short informal email to
spm-kurs@uke.de .