Applications of Unsupervised Clustering in Functional Magnetic Resonance Imaging
Type of DegreePhD Dissertation
DepartmentElectrical and Computer Engineering
Restriction TypeAuburn University Users
MetadataShow full item record
Functional Magnetic resonance imaging (fMRI) is a noninvasive neuroimaging technique that measures brain activity by detecting changes associated with blood flow. This technique has been widely used in radiology, biomedical research, and clinic diagnostics. Clustering is one of the most popular techniques for fMRI data analysis. The goal of clustering is to group objects in a way such that objects in the same cluster are similar to each other whereas objects in the different clusters are dissimilar. The commonly used k-means clustering and semi-supervised clustering methods require the number of clusters to be predefined, which is difficult to be determined in the majority of real data. Thus, in this dissertation, three unsupervised clustering methods were specifically chosen, which did not require a priori specification of the number of clusters. We investigated the feasibility of these methods in three different fMRI studies. In the first study, the selected unsupervised methods were adopted on resting-state functional magnetic resonance imaging connectivity measures to investigate whether the clinical diagnostic grouping of different disorders is grounded in underlying neurobiological and phenotypic clusters. A general analysis pipeline was derived along with three supplementary analyses, i.e., site-specific analysis, outlier subject elimination, and enrichment analysis. The effectiveness of proposed methods were verified on different disorders and the results suggest that neurobiological and phenotypic biomarkers could potentially be used as an aid by the clinician, in additional to currently available clinical diagnostic standards, to improve diagnostic precision. In the second study, we investigated the perforant pathway between entorhinal cortex and the hippocampus during encoding task by applying the selected clustering methods on the functional connectivity between cortical layer II of entorhinal cortex and hippocampus. The result showed that the functional connectivity between EC layer II and hippocampus parcellated the hippocampus into proximal and distal regions along perforant pathway. This parcellation was based on our observation of stronger connectivity between layer II of EC with hippocampal subfields such as DG/CA4/CA3/CA2 which are proximal to the EC along the perforant pathway, compared to subfields such as CA1/Subiculum which are distal. Further, this pattern was true more for the left, rather than the right, hippocampus. Our results provide the first direct non- invasive functional evidence for the perforant pathway in humans. In the third study, we employed selected clustering methods on functional connectivity between the hippocampus and different layers of the dorsal attention network and the default mode network to investigate HERNET (hippocampal encoding/retrieval and network) model, which proposed an encoding/retrieval dichotomy with the anterior hippocampus more connected to the dorsal attention network during memory encoding, and the posterior portions more connected to the default mode network during retrieval. Our results support some predictions of the HERNET model including anterior-posterior gradient along the long axis of the hippocampus. While preferential relationships between the entire hippocampus and DAN/DMN during encoding/retrieval, respectively, were observed as predicted, anterior-posterior specificity in these network relationships could not be confirmed. The strength and clarity of evidence for/against the HERNET model were superior with layer-specific data compared to conventional volume data.