Back to search

National Technical University of Athens (NTUA)

Ανάπτυξη υπολογιστών μεθόδων εκμάθησης μη γραμμικών πολλαπλοτήτων και αιτιότητας, για την αναγωγή της δυναμικής και τη διερεύνηση της διασυνδεσιμότητας μεγάλης κλίμακας δεδομένων, με έμφαση σε προβλήματα της υπολογιστικής νευροεπιστήμης

Abstract

dc:description

My Ph.D. dissertation focus on the development of computational methods for the analysis and identification of emergent spatio-temporal patterns from big data in the field of computational neuroscience. Specifically, the methods are met on the intersection of the disciplines of computational methods for data reduction, machine and manifold learning, complex network theory and computational neuroscience. In general, the proposed framework can be applied in a wide range of applications in engineering where big data either from experiments and/or from microscopic simulations are available. Thesis is divided into three main parts, which constitute the main body of our research. The first part consists of a rigorous comparative analysis of linear and non-linear manifold learning methods (MDS, ISOMAP, Kernel PCA, Diffusion Maps, LLE) for the diagnosis of schizophrenia based on graph theoretical measures in global scale. Here, we address a novel methodology on the construction of embedded functional connectivity networks (FCN) using manifold learning and in particular Diffusion Maps. That was the first time that such a framework is presented in the international literature. The performance of the proposed computational framework was compared against “traditional” approaches for the construction of FCN from fMRI data towards the identification of spatio-temporal patterns that can serve as biomarkers for diagnostic purposes. Well established machine learning methods such as Support Vector Machines (SVM), Artificial Neural Networks (ANN) and the nearest neighbor classifier were used to discriminate between patients and healthy controls. The second part of this Thesis concerns a new manifold learning methodology based on ISOMAP. By implementing this methodology, we were able to uncover spatio-temporal macroscopic differences in coarse areas of the brain using Functional Magnetic Resonance Imaging (rsfMRI) data from schizophrenic patients that followed a cognitive behavioral oriented group intervention program (specifically, the Integrated Psychological Therapy). Based on the proposed framework, we were able to monitor changes in the brain connectivity after treatment completion, thus confirming the beneficial effects that could be attributed to the therapeutic intervention. Again, such an analysis was presented for the first time in the international scientific community. We should note that such coarse-grained differences were not possible to be identified by the “classical” correlation approach that is usually used for such analyses. In the final and third part of this dissertation, we extended the previous methodology for the construction of embedded FCN investigating local graph theoretical properties while also mixing feature selection methods in the analysis. This time, we exported the signal of interest using an anatomical atlas and calculated local graph theoretic properties of the constructed networks. The purpose of that third part was to accurately identify biomarkers of the disease, which could lead to an accurate diagnosis of patients. Towards this aim, we applied popular machine learning/feature selection methods such as the minimum absolute shrinkage and selection operator (LASSO) and random forests (RF). Finally, we meticulously chose the method of evaluation and training of the final model to avoid over-fitting.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Gallos, Ioannis
  • Γάλλος, Ιωάννης

Subjects

dc:subject × 26

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/51167

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Gallos, Ioannis; Γάλλος, Ιωάννης. Ανάπτυξη υπολογιστών μεθόδων εκμάθησης μη γραμμικών πολλαπλοτήτων και αιτιότητας, για την αναγωγή της δυναμικής και τη διερεύνηση της διασυνδεσιμότητας μεγάλης κλίμακας δεδομένων, με έμφαση σε προβλήματα της υπολογιστικής νευροεπιστήμης. National Technical University of Athens (NTUA), 2022. http://hdl.handle.net/10442/hedi/51167