Functional brain network identification in obesity using resting-state fMRI BOLD signals analysis

Abstract

Identifying altered functional brain networks is crucial for unraveling the neurobiological mechanisms underlying obesity. Resting-state functional magnetic resonance imaging (rs-fMRI) provides a noninvasive means to characterize spontaneous neural activity and map large-scale functional brain networks; however, conventional analyses often overlook the dynamic time-frequency patterns of blood oxygenation level-dependent (BOLD) signals. This study, for the first time in the literature, employs data-driven machine learning (ML) to analyze time-frequency patterns of local neural activity within key functional regions and identify networks that differentiate subjects with obesity from normal-weight healthy controls (HC). Using rs-fMRI data from the Human Connectome Project (HCP), we extracted time-frequency features from three well-established networks, i.e., the Default Mode Network (DMN), Salience Network (SN), and Executive Control Network (ECN), collectively known as the triple-network model, which plays a central role in regulating cognitive control, self-referential processing, and reward evaluation during eating behaviors, for 141 participants with obesity and 141 HC participants. Each network was analyzed independently to determine optimal feature selection strategies and model architectures for accurate classification. Our results demonstrate that ML-based time-frequency features achieve the strongest classification performance, as measured by area under the ROC curve (AUC) for the left-to-right (LR) phase-encoding direction: ECN (0.7136), SN (0.7121), and DMN (0.6102). These findings highlight that the DMN, SN, and ECN feature profiles identified through our ML-driven time-frequency analysis could serve as novel biomarkers for obesity, demonstrating the promise of our approach for clinical neuroimaging research.