We are working on a Matlab-based toolbox to facilitate multi-voxel pattern analysis of fMRI neuroimaging data. The toolbox is currently in beta testing, with the first public release planned for later this year.
The aim of the MVPA toolbox effort is to create a set of open-source functions in a widely used language to facilitate exploration of multi-voxel pattern analysis techniques and to reduce the 'startup costs' for knowledgeable users eager to apply pattern classification algorithms to their imaging data. By developing the toolbox in the Matlab environment, users are able to take advantage of the vast array of existing functions. The data structures used and generated by the toolbox are designed to facilitate exploration and further script development.
You can join or read the archives of the mvpa-toolbox@googlegroups.com public discussion mailing list at http://groups.google.com/group/mvpa-toolbox.
The latest version of MVPA has officially been released, this release is a combination bug-fix and general maintenance release. See the Google Groups mailinglist for more information on the bug, which is related to the train_logreg.m file. Thanks to Jesse Rissman for finding this bug and reporting it.
The latest svn version of the load/write spm file handlers now feature Nifti support. Nifti files can be opened in the same manner as Analyze files up till this point and the write_to_spm function now defaults to writing out a Nifti file with the ability to override the current file format using the optional arguement 'FEXTENSION'.
The unit tests have been updated to reflect this and new versions of the data have been posted on the download page. Each of the data sets now includes everything you need to run the tutorial_easy for that data format and the associated unit tests. Note: to run the unit tests for the Nifti/Analyze data you will also need the Afni data set.
We are proud to announce version 1.0 of the MVPA Toolset. The link below will allow you to download the newest official release of the software, and of course you can always keep yourself up to date using the SVN links in our wiki. Please see the file progress/changelog_1.0.txt for a comprehensive list of all of the changes.
From now on, we'll be using mvpa-toolbox@googlegroups.com as our public discussion list. Feel free to join or browse the archives.
Thomas Wolbers was nice enough to send us a set a high pass filter script and detrending script. They have been officially added to the mvpa SVN. This new functionality can be found in core/preproc and the file names are hpfilter_runs.m and detrend_runs.m. You will need the SPM library to make hpfilter_runs.m function (it relies upon spm_filter.m).
From now on, the Subversion development repository containing the latest versions of all the core MVPA toolbox scripts will be publicly available. If you want to live on the bleeding edge, or need functionality that hasn't made its way into one of the official releases, you can grab anything you're missing from here. For information on what a Subversion version control system is and how to access it, see the wiki page.
We have officially released the beta scripts for importing and exporting ANALYZE files (e.g. from SPM) directly into/out of the MVPA toolbox. These rely on SPM's import/export scripts, wrapped to make use of MVPA toolbox data structures.
We've done our best to test them carefully, but we advise users to read the documentation about them before use, to pay close attention to what they produce, and to let us know if anything seems awry.
You can access them from the Subversion repository or as a separate download (see the 'Download' section below). Data for testing the scripts has been included as well.
If you like the idea of multi-voxel pattern analysis, but don't or can't use Matlab, then you may be interested in PyMVPA. This is a Python-based toolbox similar in spirit to our Matlab MVPA toolbox, written by Michael Hanke, Yaroslav Halchenko, Per Sederberg and the rest of the Debian Experimental Psychology crew.
Let us know if you have any problems!
The 2007 Pittsburgh (EBC) brain activity interpretation competition is kicking off. In the webcast, Walt Schneider and Greg Siegle said that the data will be provided in a standard Matlab format, and also in an MVPA toolbox 'subj' structure.
The tutorial dataset for the EBC Extension has been released. You can now download everything you need to run the tutorial and generate EBC submissions for movies 1 and 2. Furthermore, as of MVPA version 0.8, the EBC Competition Extension is included as part of the standard MVPA releases. You will no longer need to install the EBC toolbox separately.
Please see the EBC Extension page for more information.
Version 0.8 of the MVPA toolbox beta has been released, and is now available for download. It should be fully backwards compatible with previous releases. This is also the first release to use our new wiki documentation system - bear with us while we straighten this out, and feel free to get involved in editing. Just create a login for yourself, and away you go. For a more detailed list of the changes between 0.8 and 0.7.1, please see the changelog.
A beta version of the EBC Competition extension for the MVPA toolbox has been released. This release contains all of the optimizations used in the Princeton EBC Team's prize winning entry, so that users may replicate their results. Please see the EBC Extension page for more information.
Important disclaimer: At this point in time, the Princeton MVPA toolbox is unsupported beta software, which we are making available to anyone who might find it useful. Because the software is still in beta form, there may be bugs and othe teething problems. We do not take any responsibility for any problems that you might have related to use of the software. If you find a bug or have any further suggestions, you should let us know, but (given the presently unsupported status of the software), we may not be able to reply to all of the queries that we receive regarding the software.
If you are interested in downloading a beta version of the toolbox, please download:
Having downloaded them, follow these instructions, and then you should be ready to start the tutorial. Contact us if you have trouble with the installation.
As of release 0.8, setup instructions, the manual, the tutorials, and the glossary can all be found at the MVPA Documentation Wiki. Quick links are provided below.
The MVPA toolbox has been described in posters presented at the Annual Meeting of the Organization of Human Brain Mapping:
The Multi-Voxel Pattern Analysis (MVPA) toolbox - abstract
(2006)
(poster #50, presented on Wed June 15th in the afternoon)
Greg J Detre, Sean M Polyn, Christopher D Moore, Vaidehi S Natu,
Benjamin D Singer, Jonathan D Cohen, James V Haxby, Kenneth A Norman
A
Matlab-based toolbox to facilitate multi-voxel pattern classification of fMRI
data (2005)
Sean M Polyn, Greg J Detre, Sylvain Takerkart, Vaidehi S Natu, Michael
S Benharrosh, Benjamin D Singer, Jonathan D Cohen, James V Haxby,
Kenneth A Norman
See the Princeton EBC page for more information on the EBC 2006 competition.
Note: We highly recommend upgrading to the latest version, as each upgrade is mostly backwards-compatible.
Center for the Study of Brain, Mind and Behavior