Browsing by Subject "MVPA"

The goal of this thesis was to examine how the representation of faces and identities has been researched with multivariate brain-imaging analysis methods. The coding of the visual identity of faces is a complex process, which arises from abstract high-level representations that are viewpoint independent. This process has been localized to certain visual brain areas, namely FFA and ATL-FA in humans and AM in macaque monkeys. The coding processes of identities have been researched with functional magnetic resonance imaging (fMRI) and multivariate pattern analysis (MVPA) methods, where the measured brain activity distribution is compared with the distributions predicted by the processing models of identity. Most models are based on face spaces, which are defined by the identities, features or abstract properties of a face. The neural basis for coding of the identity of faces has not yet been resolved. However, the low-level and feature models were best at predicting the data, although low-level models explain mainly the properties of the stimuli and the function of V1. The prototype model was not as successful in predicting the activity distributions as the other models. In general, the models explained the orientation of the faces better than their identity. This might be due to the differences between the brain activities associated with different identities being so subtle that the multivariate pattern analysis cannot differentiate between them. It is also possible that the coding of identity does not happen within the face processing areas, but perhaps in the connections between frontal and face areas. To conclude, the usefulness of fMRI-MVPA in studying identity has been questioned and the methods still need improvement. However, MVPA is a more versatile and sensitive method of investigating the coding of identity than traditional univariate analysis methods.

Visual working memory is a cognitive system that is responsible of short term storage and manipulation of visual information. Working memory is divided to memorising, storage and recalling of the stimulus. This review concentrates in visual working memory studies that used functional magnetic resonance imaging (fMRI). FMRI is a spatially accurate and is based on changes in the brains blood circulation. The data from fMRI can be analysed with univariate or multivariate methods. These methods answer different research problems because of their different premises. The premise of univariate analysis is that the neural activation in one part of the brain is directly related to its function. In multivariate analysis, the neural activation is approached by observing the activations distribution, which means that different activation distributions in same parts of the brain can be related to different processes. The visual areas of the brain are located in the occipital lobe but, before multivariate analysis, the visual working memory has been associated with prefrontal cortex. After multivariate pattern analysis (MVPA) has increased in popularity the hypothesis, that visual areas have a part in visual working memory, has also generalised. Because of the activation distribution premise, the MVPA is a more sensitive method to analyse fMRI data. Still there have been different results in different MVPA using studies. Different memory tasks might also be partly responsible of different results. A visual working memory task always activates prefrontal and parietal cortices in addition to sensory cortex. Visual cortex seems to have the principal part and prefrontal and parietal cortices take part most likely in executive functions but they can’t be ruled out from storage either.