Beyond decoding: representational similarity analysis on fMRI data
Name
Martin Loginov
Abstract
Representational similarity analysis is a novel data analysis technique in neuroscience
first proposed by Kriegeskorte et al. in [KMB08]. It aims to connect different branches
of neuroscience by providing a framework for comparing activity patterns in the brain
that represent some cognitive processes. These activity patterns can come from various
sources, like different subjects, species or modalities like electroencephalography (EEG)
or functional magnetic resonance imaging (fMRI). The central concept of RSA lies in
measuring the similarity between these activity patterns. One of the open questions
regarding RSA is what distance measures are best suited for measuring the similarity
between activation patterns in neuronal or fMRI data.
In this thesis RSA is implemented on a well known fMRI dataset in neuroscience,
that was produced by a studying the categorical representations of objects in the ventral
temporal cortex of human subjects [HGF + 01]. We carry out RSA on this dataset using
different notions of distance and give an overview of how the end results of the analysis are
affected by each distance notion. In total 9 different distance measures were evaluated for
calculating the similarity between activation patterns in fMRI data. The results provided
in this thesis can be used by researchers leveraging RSA to select distance measures for
their studies that are most relevant to their particular research questions at hand. In
addition to the comparison of distance notions, we also present a novel use case for RSA
as a tool to visualize the global effects different transformations can have on the input
dataset.
Graduation Thesis language
English
Graduation Thesis type
Master - Information Technology
Supervisor(s)
Raul Vicente Zafra
Defence year
2015