Over the past decade, pattern classification methods have become widespread in functional magnetic resonance imaging (fMRI).
These methods, typically referred to as multivoxel pattern analysis (MVPA) or multivariate pattern decoding, are now applied
to a wide range of neuroscientific questions. There has been particular interest in applying these approaches, e.g., in detecting
deception or for diagnostic purposes. In this review, we will focus on what can be achieved by pattern classification analyses
of fMRI data; the strengths and weaknesses of this approach; and the biological processes giving rise to the signals measured
by this method. Finally, we will discuss how these multivariate approaches are starting to be applied to the analysis of anatomical
magnetic resonance imaging (MRI) and magnetoencephalographic (MEG) data. WIREs Cogni Sci 2011 2 568–579 DOI: 10.1002/wcs.141
Figure 1.
Illustration of multivoxel pattern information. The scatter plots show repeated measurements from two experimental conditions (denoted by red and blue dots). The two axes reflect the response measured in two voxels. A nonlinear classifier (a) can discriminate the conditions perfectly in the training data. However, when additional independent test measurements are made (black dots), the classifier fails to generalize because it models the noise in the data. Instead, a linear classifier (b) allows for some misclassification in the training data, but may generalize well to independent test data.
Pattern classification and functional magnetic resonance imaging (fMRI) have been used to track the subjective perceptual state over (Reprinted with permission from Ref 3. Copyright 2005 Cell Press) (a) and to show which visual stimulus participants held in working (Reprinted with permission from Ref 14. Copyright 2009 Nature Publishing Group) (b). Further, it was employed to decode a person's location within a virtual reality environment from responses in the hippocampus (Reprinted with permission from Ref 15. Copyright 2009 Cell Press) (c). It is even possible to use this approach to reconstruct a visual stimulus seen by a participant (Reprinted with permission from Ref 13. Copyright 2008 Cell Press) (d).
The biological basis of voxel patterns. (a) The hyperacuity hypothesis assumes that the coarse voxel grid leads to biased sampling of the fine-grained underlying cortical architecture (top). Because a particular voxel in early visual cortex may contain a greater number of neurons responding to one orientation than others (see color code), the response to this orientation may be slightly greater (bottom). (Reprinted with permission from Ref 37. Copyright 2005 Nature Publishing Group) (b) Filtering the data with different bandwidths shows that voxel bias patterns for orientation in visual cortex exist at multiple scales. (Reprinted with permission from Ref 40. Copyright 2010 Society for Neuroscience) (c) A complex spatiotemporal filter model can explain how anisotropic cortical architecture and neurovascular coupling give rise to voxel response biases. (Reprinted with permission from Ref 41. Copyright 2010 Elsevier)
David Alais is Associate Professor in the School of Psychology at the University of Sydney, Australia. He has been with the University of Sydney since 2003, before which he was a research fellow in France, funded by the Human Frontiers Science Programme at the Collège de France in Paris, and subsequently a Marie Curie fellow at the Neuroscience Institute in Pisa, Italy.
Professor Alais’ current research aims to better understand visual and auditory perception. One major area of study uses human psychophysical experiments to better understand how the brain combines auditory and visual information to enhance our perception of the world.
