Node Embedding Extraction for Causal Brain Graphs in fMRI Data

This study introduces a novel pipeline for feature extraction from Functional Magnetic Resonance Imaging (fMRI) data, focusing on age-related trends in node embeddings derived from causal brain graphs during naturalistic stimuli (movie observation). Directed Acyclic Graphs (DAGs) constructed using the Non-combinatorial Optimization via Trace Exponential and Augmented Lagrangian for Structure learning (NOTEARS) algorithm modeled causal relationships among 25 brain regions of interest (ROIs). Graph embeddings were generated using the Deep Walk algorithm to capture low-dimensional representations of graph structures. Unlike previous studies analyzing functional brain graphs with traditional graph metrics (e.g., degree, centrality, clustering coefficient), our approach employs modern techniques to uncover patterns in effective brain graphs. The results revealed decreasing feature similarity with age in regions associated with Theory of Mind (ToM), Pain, Face and Place Processing, reflecting age-related neural changes. Conversely, the Left Retrosplenial Cortex (IRSC) demonstrated increasing similarity, suggesting enhanced maturation and efficiency in spatial memory functions. Regions like the Left Fusiform Face Area (IFF A) and Right Transverse Occipital Sulcus (rTOS) showed no significant trends, emphasizing the role of functional relevance in brain responses to naturalistic stimuli. The novelty of our research lies in analyzing localized patterns in causal brain graphs using embedding techniques like Deep Walk, moving beyond traditional metrics. For the first time, features were extracted from fMRI data to study node-level similarities across age groups during naturalistic stimuli. By combining effective graph construction with localized embedding methods, this pipeline advances age-related brain connectivity research, demonstrating the consistent impact of movie stimuli on brain dynamics and the potential of graph-based methods to reveal age-related neural connectivity changes.