Multi-Voxel Pattern Analysis
Multi-Voxel Pattern Analysis
Multi-Voxel Pattern Analysis is a way to analyze brain scans obtained using fMRI. In contrast to fMRI analysis which focuses on individual brain voxels (volumetric pixels), MVPA utilizes pattern-classification algorithms to figure out the information that is represented in a specific brain activity pattern. The benefits to using MVPA over traditional single voxel analysis is that MVPA is a more sensitive analytical tool. In particular, it is more useful for hypothesis that think of cognitive states as a result of activity in many interacting parts of the brain rather than a centralized location. Another important benefit of the MVPA approach is the ability to correlate classifier estimates made by the algorithm to the subject's behavior. Thus this technique allows us to sensitively detect and track cognitive states as well as see how these states are represented in the brain. 1Additionally, it can be used to interpret overlapping functional activations.2
One issue that plagues fMRI studies is how to interpret overlapping activity from two or more contrasts of experimental conditions. Generally, two different interpretations can be used. Either the shared area of activity could be interpreted as having common computational process, such as in brain areas that are activated by both observed and performed manual actions, or the shared area of activity could be thought of to have two functionally independent neural populations that happen to operate in the same region. The former interpretation has been used to justify the existence of mirror neuron systems in the human brain. MVPA could be used to show that the similar patterns of activation between different environmental conditions are unrelated.2
- Feature Selection: Decide which voxels to include in the classification analysis.
- Pattern Assembly: Sort data into discrete "brain patterns" corresponding to pattern of activity across selected voxels at a particular time in the experiment. Label brain patterns activity to the experiment while accounting for delays in scanners.
- Classifier Training: give a subset of labeled patterns to a multivariate pattern classification algorithm. This allows the algorithm to figure out a formula that relates the experimental condition and the voxel activity patterns.
- Generalization Testing Give the classification algorithm a new pattern of brain activity that it hasn't seen. and see if the it is able to determine the experimental condition associated with the pattern.
Multivariate methods have been used to analyze fMRI data in order to figure out relationships between different brain regions. In the last 5 years, scientists have realized that these same principles could be used to characterize cognitive states as patterns of brain activity. This is advantageous because many complex cognitive states have widely distributed brain activity patterns; for example, the activity pattern associated with thinking about a nail may elicit brain activity in more than just, say, the visual cortex. So determining cognitive states from a brain activity pattern can not be associated to only one area of activity.
1 Norman, K.A. et. al.Beyond mind-reading: multi-voxel pattern analysis of fMRI data.TRENDS in Cognitive Sciences Vol. 10 No.9. pg. 424-430.
2 Peelen, M.V. and Paul E. Downing. Using Multi-voxel pattern analysis of fMRI data to interpret overlapping functional activations. Trends in Cognitive Sciences Vol.11 No. 1 pg. 4-5.
- Task or test associated with this construct (vote for your favorite, or nominate a new one by editing this page)
- Indicators (dependent variables, conditions, or contrasts; measurement variables used for analysis) associated with this construct (vote or nominate by editing this page):
- Closely related pages (vote or nominate related pages by editing this page):
Hanke, Micael et. al. PyMVPA: a Python Toolbox for Multivariate Pattern Analysis of fMRI Data. Neuroinform (2009) 7:37-53.
- CNP Level (What's this?)
- Cognitive Concept
- Links out:
- Database links