VitorCarvalho_Thesis.pdf
Functional connectome (FC) fingerprinting is a rapidly evolving field within neuroimaging that seeks to identify unique patterns of functional connectivity in the human brain. The ability to reliably distinguish individuals based on their functional connectome has significant implications for neuroscience, particularly in understanding cognitive variability, mental disorders, and personalized medicine. Traditional approaches to FC fingerprinting rely on direct comparisons of functional connectivity matrices, but these methods can be limited by noise, high dimensionality, and the inherent variability of fMRI data across scanning sessions. To address these challenges, this thesis explores two distinct approaches utilizing dimensionality reduction techniques to enhance FC fingerprinting.
The first chapter investigates the use of tensor decomposition to improve the identifiability of individual functional connectivity patterns across multiple fMRI conditions. By decomposing high-dimensional FCs data into a core tensor and factor matrices, the framework successfully enhances the stability and distinctiveness of functional connectome fingerprints. The study examines within-condition and between-condition fingerprinting, demonstrating that Tucker decomposition significantly increases matching rates compared to methods that do not model the high-dimensionality of functional connectivity data, particularly for lower parcellation granularities.
The second chapter focuses on the application of Riemannian geometry-based methods, particularly tangent space projection, to further refine functional connectome fingerprints when several data acquisition sessions are available. This approach addresses the limitations of traditional Euclidean-based analyses by incorporating the geometric space in which FCs lie mapping FC matrices onto a tangent space, enabling robust comparisons while preserving individual-specific patterns. By leveraging Pearson correlation and Euclidean distance metrics, the study highlights the advantages of tangent FCs in improving subject distinctiveness across multiple scanning sessions. The integration of t-SNE visualization further validates the effectiveness of this method in clustering subject-specific connectivity patterns.
History
Degree Type
- Master of Science
Department
- Industrial Engineering
Campus location
- West Lafayette