University of Crete (UOC)
Graph signal processing and recurrence quantification techniques for the analysis of biomedical signal ensembles
Abstract
dc:descriptionThis thesis develops novel mathematical signal processing methods for the analysis of biomedical signals. We address three distinct biomedical applications. First, we extend graph signal processing theory using fractional lower-order statistics in order to robustly denoise electroencephalogram (EEG) signals corrupted by additive heavy-tailed impulsive noise modeled with alpha-stable distributions. The proposed graph-based filter adapts to the statistics of the impulsive noise, due to head motions or the environment, and it suppresses its effects on EEG signals.The second part concerns the nonlinear dynamic analysis of resting-state functional magnetic resonance imaging (rs-fMRI) time series. As the human brain is highly dynamic mathematical methods should approach the human brain functionality in a dynamical manner. Recurrence is a fundamental property of dynamical systems, which can be exploited to characterize the system's behavior in phase space. We employ cross recurrence quantification analysis (CRQA) to exploit the recurrence plots and via their quantitative analysis extract specific features, which characterize the dynamic behavior of rs-fMRI recordings. We showed that this method is able to recognize regional brain interconnections allowing us to separate healthy controls from patients with neuropsychiatric systemic lupus erythematosus (NPSLE). We extended our approach to create brain networks through the linear combination of the individual CRQA-based features. The analysis of these networks through conventional graph theory of small-world led to a high classification accuracy.Finally, we addressed the problem of accelerating Diffusion-Weighted MRI (DW-MRI). DW-MRI is a non-invasive technique that provides information about the microstructure of human organs via long duration examinations. Decreasing DW-MRI examination time reduces patient discomfort. In DW-MRI, b-value is a factor that reflects the strength and timing of the gradients used to generate diffusion-weighted images. The higher the b-value, the stronger the diffusion effects. Through the application of a sparse representation method, we sampled a limited number of b-values and we reconstructed the rest through trained dictionaries. Experimental results on pancreas imaging showed the potential of our approach.
Degree
thesis:*- Grantor dc:publisher
- University of Crete (UOC)
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Pentari, Anastasia
- Πεντάρη, Αναστασία
Subjects
dc:subject × 31- Επεξεργασία σήματος μέσω γράφων
- Ηλεκτροεγκεφαλογράφημα
- Λειτουργική συνδεσιμότητα εγκεφάλου
- Τοπολογική συνδεσιμότητα εγκεφάλου
- Αναδρομική ποσοτική ανάλυση ζεύγους
- Λειτουργικός μαγνητικός τομογράφος
- Δυναμική λειτουργική συνδεσιμότητα
- Δυναμική λειτουργία του εγκεφάλου
- Δίκτυα εγκεφάλου
- Ανάλυση μικρού-κόσμου
- Αραιές αναπαραστάσεις
- Μαγνητική απεικόνιση σταθμισμένης διάχυσης
- παράμετρος β
- Graph signal processing
- EEGs
- Functional connectivity
- Topological connectivity
- Cross recurrence quantification analysis
- rs-fMRI
- Dynamic brain functionality
- Brain networks
- Small-world analysis
- Sparse representations
- DW-MRI
- b-value
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Ηλεκτρική και Ηλεκτρονική μηχανική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Electrical and Electronic Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/52157