|Reference : Multivariate pattern analysis: brain decoding|
|Parts of books : Contribution to collective works|
|Engineering, computing & technology : Electrical & electronics engineering|
Human health sciences : Neurology
Human health sciences : Multidisciplinary, general & others
|Multivariate pattern analysis: brain decoding|
|Schrouff, Jessica [Université de Liège - ULg > > Centre de recherches du cyclotron >]|
|Phillips, Christophe [Université de Liège - ULg > > Centre de recherches du cyclotron >]|
|Coma and altered states of consciousness|
|[en] Machine learning ; Neuroimaging data ; Introduction|
|[en] Two of the most fundamental questions in the field of neurosciences are how information is represented in different brain structures, and how this information evolves over time. Various tools, such as Magnetic Resonance (MRI) and Positron Emission Tomography (PET) have been developed over the last few decades to record brain activity and investigate these questions. In particular, functional MRI (fMRI) tracks changes of the Blood Oxygenation Level-Dependent (BOLD) signal, which is a good indicator of brain activity, with a spatial resolution of a few cubic millimeters and a typical temporal resolution in the order of 1 or 2 seconds.
Until recently, the methods used to analyze such data focused on characterizing the individual relationship between a cognitive or perceptual state and each image voxel, i.e. following a massively univariate approach. A well-known univariate technique is Statistical Parametric Mapping (SPM) . SPM relies on the General Linear Model to detect which voxels show a statistically significant response to the (combination of) experimental conditions of interest. However, there are limitations on what can be learned about the representation of information by examining voxels in a univariate fashion. For instance, spatially distributed sets of voxels considered as non-significant by a SPM analysis of one experimental condition might still carry information about the presence or absence of that condition. Furthermore, classic voxel-based analytic techniques are agnostic of any a priori information, for example disease-specific information. They are also mainly designed to perform group-wise comparisons and would therefore be unsuitable to evaluate the state of the disease of each individual.
On the other hand, Multi-Voxel Pattern Analyses (MVPA) allow an increased sensitivity to detect the presence of a particular mental representation. These multivariate methods, also known as brain decoding or mind reading, attempt to link a particular cognitive, behavioral or perceptual state to specific patterns of voxels’ activity. Application of these methods made it possible to decode the category of a seen object or the orientation of a stripped pattern seen by the subject from the brain activation of the imaged subject. Advances in pattern-classification algorithms also allowed the decoding of less-controlled conditions such as memory retrieval tasks. Advanced mathematical tools are still under development to allow the classification of more complicated experimental data sets, such as examining the content of mind wandering or detecting the state of consciousness of a patient showing no response to a command.
|Centre de Recherches du Cyclotron - CRC ; Coma Science Group|
|Fonds de la Recherche Scientifique (Communauté française de Belgique) - F.R.S.-FNRS ; Fonds pour la formation à la Recherche dans l'Industrie et dans l'Agriculture (Communauté française de Belgique) - FRIA|
|Researchers ; Professionals ; Students|
|File(s) associated to this reference|
All documents in ORBi are protected by a user license.