Multi-voxel pattern analysis
Traditional neuroimaging analysis techniques are designed to detect the "activation" of neural structures that are at least a centimetre or so in scale. However, a growing number of studies have shown that substantial information can be obtained from the fMRI signal at a scale that is smaller than this. To do this, pattern classifiers are used to relate distinct patterns of activation within a brain region to corresponding mental states (e.g., stimuli). Although the pattern is generally idiosyncratic in that it varies from one person to the next, within subjects, reliable correspondences can be found.
MVPA has been shown to reveal information within brain regions that was lost in previous fMRI analyses, such as sensitivity to orientation in V1.
Many different classifiers have been used to build the correspondence between activity pattern and mental state, including correlation, linear-discriminant analysis and support vector machines. In some studies, just one a priori region-of-interest is examined, in others the whole brain classified (so providing no specificity), and in others, a roving "searchlight" is used to examine the whole brain one region at a time.
There is a weekly Representational Similarity Analysis Interests Group (RSAIG) meeting to discuss method development and applications of MVPA. Topics have covered in the past related to Multivariate Pattern Analysis (MVPA). For instance, Representational Similarity Analysis (RSA) as the list name suggests and various pattern classification approaches to fMRI and E/MEG analysis. For more detail of this meeting, please contact the group coordinator Li Su. To subscribe or unsubscribe the mailing list, please visit: http://lists.mrc-cbu.cam.ac.uk/mailman/listinfo/rsaig. The mailing list hosts more than 30 members of the Unit and researchers in the University. It publish the topics of each meeting and circulate information on the RSA toolbox.
There are also other periodic meetings at the CBU to discuss this topic. See MriPhysicsAndAnalysisForCognitiveNeuroscientists
Representational Similarity Analysis
RSA is a particular versatile extension of MVPA.It goes beyond testing for information in regional response patterns and enables us to handle condition-rich experiments without predefined stimulus categories, to test conceptual and computational models, and to relate representations between individuals and species (Kriegeskorte et al. 2008a, 2008b). Here we present a Matlab toolbox for representational similarity analysis. RSA characterizes the representation in each brain region by a representational dissimilarity matrix (RDM). An RDM is a square symmetric matrix, each entry referring to the dissimilarity between the activity patterns associated with two stimuli (or experimental conditions).
RSA toolbox developed in CBU
A group of researchers in CBU have developed a Matlab toolbox for representational similarity analysis. This toolbox, which is the first toolbox to implement RSA, is modular and work-flow based. There are a set of "Recipe" functions in the toolbox that allow automatic ROI analysis as well as whole-brain searchlight analysis. Each recipe is formed of a number of "modules". Each module performs one set of analysis e.g. one of the modules displays the representational geometries in a specific region of interest. The analysis steps as implemented in the toolbox are the following: step1: Calculating and visualizing brain RDMs Step2: comparing different RDMs Step3: statistical inference by bootstrap and randomisation tests
In addition to the different modules and recipes,etc. the toolbox also allows the users to simulate arbitrary regional pattern ensembles. They can then apply the toolbox modules to the simulated data and compare the results with the simulated ground truth. This would both make them familiar with the structure of the codes and also allow testing different hypothesis, etc. on the simulated data.
The figure bellow shows the representational geometries obtained by applying the toolbox module that displays the representational geometries for noise-less and noisy fMRI data.