Vis enkel innførsel

dc.contributor.advisorLind, Pedro
dc.contributor.advisorMirtaheri, Peyman
dc.contributor.authorRomero, Sergio Alejandro Sotres
dc.date.accessioned2021-09-09T08:50:15Z
dc.date.available2021-09-09T08:50:15Z
dc.date.issued2021
dc.identifier.urihttps://hdl.handle.net/11250/2774866
dc.description.abstractThe functional near-infrared spectroscopy (fNIRS), as a brain imaging modality, is a versatile technique for understanding brain activity processes at the level of the brain cortex. The use of this technology facilities the understanding of brain metabolism, oxygenation, and its related brain activity parameters when participants perform dynamical tasks. In this thesis, we apply different methods to extract the functional connectivity network of the brain, employing data generated by this technology. This functional connectivity network is a measure that qualitatively informs the interconnection of different regions of the brain. To perform such a task, we calculated the Pearson correlation coefficients and the mutual information between pairs of signals from fNIRS data, to determine the strength of shared information among them. We construct weighted networks that display the more correlated regions and compare these methods to unsupervised learning techniques such as PCA, ICA, and dendrograms. Additionally, we include an implementation where we explore nonlinear dependencies of fNIRS data using mutual information. From this analysis, we observed that a mutual information approach based on binning techniques allows quantifying more general correlations than using the Pearson coefficient but is highly susceptible to bias. The method also provides more relevant information compared to the PCA and ICA, since with the last one, we can observe the dependencies of signals but in a disorderly manner. The resulted bias is been reflected in lower values that are more visible when doing a threshold examination (\ref{comparison},\ref{numedges}). A deeper analysis in this regard to bias reduction needs further exploration in future work. Additionally, the calculation of a coefficient (referred to in the thesis as $\Lambda$) that distinguishes the type of dependence between random variables resulted to be a useful method for fNIRS data. Such a coefficient indicates a clear way to quantify linear and nonlinear dependencies by using mutual information, but with the incapability of reflecting the specific type of behavior involved.en_US
dc.language.isoengen_US
dc.publisherOsloMet - storbyuniversiteteten_US
dc.relation.ispartofseriesACIT;2021
dc.titleA mutual information approach on fNIRS functional connectivity networken_US
dc.typeMaster thesisen_US
dc.description.versionpublishedVersionen_US


Tilhørende fil(er)

Thumbnail

Denne innførselen finnes i følgende samling(er)

Vis enkel innførsel