H. A. Hernández Roig, M. C. Aguilera-Morillo, E. Arnone, R. E. Lillo Pascual, L. M. Sangalli
In this talk, we tackle the problem of discriminating between individuals with schizophrenia and healthy patients based on their Functional Connectivity (FC) maps. These FC maps were obtained from task-based functional magnetic resonance images (fMRI) and are publicly available thanks to the Consortium for Neuropsychiatric Phenomics at UCLA. We view the brain as a three-dimensional (3D) domain with a non-Euclidean structure and consider the FC maps as functional data observed over this complex domain. The discrimination task is approached by formulating a scalar-on-function (SOF) regression problem using functional partial least squares (FPLS). To solve this problem, we propose a new penalized FPLS approach based on a rank-one approximation of the data, which we refer to as R1-FPLS. Additionally, we discuss the advantages of R1-FPLS compared to other state-of-the-art approaches.
Keywords: Functional Data Analysis, Partial Least Squares, Finite Elements, Functional Connectivity, Scalar-on-function Regression
Scheduled
GT01.FDA2 Invited Session
November 8, 2023 4:00 PM
CC1: Audience