Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Multi-subject resting-state fMRI data analysis via generalized Canonical Correlation Analysis

Karakasis Paris, Liavas Athanasios, Sidiropoulos Nikos, Simos Panagiotis G., Papadaki Efrosini

Full record


URI: http://purl.tuc.gr/dl/dias/10C41380-B4B0-469B-9AEB-430BA08480DE
Year 2021
Type of Item Conference Publication
License
Details
Bibliographic Citation P. A. Karakasis, A. P. Liavas, N. D. Sidiropoulos, P. G. Simos and E. Papadaki, "Multi-subject resting-state fMRI data analysis via generalized Canonical Correlation Analysis," in 2020 28th European Signal Processing Conference (EUSIPCO), Amsterdam, Netherlands, 2021, pp. 1040-1044, doi: 10.23919/Eusipco47968.2020.9287655. https://doi.org/10.23919/Eusipco47968.2020.9287655
Appears in Collections

Summary

Functional magnetic resonance imaging (fMRI) is one of the most widespread methods for studying the functionality of the brain. Even at rest, the Blood Oxygen Level Dependent (BOLD) signal reflects systematic fluctuations in the regional brain activity that are attributed to the existence of resting-state brain networks. In many studies, it is assumed that these networks have a common spatially non-overlapping manifestation across subjects, defining a common brain parcellation. In this work, we propose an fMRI data generating model that captures the existence of the common brain parcellation and present a procedure for its estimation. At first, we employ generalized Canonical Correlation Analysis (gCCA) - a well-known statistical method, which can be used for the estimation of a common linear subspace - and recover the subspace that is associated with the common brain parcellation. Then, we obtain an estimate of the common whole-brain parcellation map by solving a semi-orthogonal nonnegative matrix factorization (s-ONMF) problem. We test our theoretical results using both synthetic and real-world fMRI data. Our experimental findings corroborate our theoretical results, rendering our approach a very competitive candidate for multi-subject resting-state whole-brain parcellation.

Services

Statistics