Banca de QUALIFICAÇÃO: BRENO CORDEIRO BISPO

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : BRENO CORDEIRO BISPO
DATE: 17/11/2023
LOCAL: Online (fechada ao público)
TITLE:

HYPERGRAPH SIGNAL PROCESSING: AN APPLICATION TO BRAIN FUNCTIONAL CONNECTIVITY ANALYSIS.

 

KEY WORDS:

Graph signal processing, hypergraph signal processing, graph learning, brain functional connectivity.

PAGES: 101
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

Advances in neuroimaging technologies, such as functional magnetic resonance imag- ing, have provided fundamental information regarding anatomical and functional interactions between different regions of the brain. On the other hand, graph theory is widely used to model irregular and non-uniform structures designed by dyadic relationships between network objects. Furthermore, a theoretical framework tailed to study the dynamics of multivariate data emerged in the last decade that aim to generalize classical signal processing techniques from regular to irregular domains, the graph signal processing. The application of these concepts in neuroscience has provided new ways of analyzing neurological behaviors and brain diseases. However, the characterization of such complex system, which naturally consists of non-dyadic interactions, through graphs has become a shortcoming. On the other hand, recent studies, which cover methodologies to model brain networks based on high-order interactions, are able to reveal hidden properties from traditional graph approaches. In this context, this work aims to apply the concepts of an emerging mathematical tool called hypergraph signal processing (considered a generalization of its version defined on graph that aims to overcome limitations of this latest version regarding the study of dynamics of high-order multivariate complex systems) in new interpretations of structural and functional brain connectivity. For this purpose, traditional and modern clustering methods were implemented in order to provide quantitative and qualitative comparisons of neurological pattern detection performance. Within the outcomes, the techniques grounded in structural and spectral analysis of hypergraph outperform graph-based methods, as evidenced by their superior performance regarding the silhouette coefficients of clusters, their ability to enhance community segregation, and their proficiency to detect brain functional features that remain elusive to graph-based approaches. In this way, it is believed that the applica- tion of hypergraph signal processing tools combined with modern machine learning techniques can provide new discoveries regarding the detection / classification of neurological diseases or promote a better understanding of brain functional connectivity signatures.

COMMITTEE MEMBERS:
Externo à Instituição - ALCEBÍADES DAL COL JÚNIOR - UFES
Presidente - 1882484 - JULIANO BANDEIRA LIMA
Externo ao Programa - 1426994 - NIVALDO ANTONIO PORTELA DE VASCONCELOS - null