🤖 AI Summary
Existing functional connectivity (FC) estimation methods—such as region-averaged correlation—are systematically biased by intra-regional correlations and measurement noise. To address this, we propose the first voxel-level FC estimation framework grounded in a linear mixed-effects model (LMM), explicitly modeling both inter- and intra-regional BOLD signal correlations alongside structured noise, with subject-specific inter-regional correlation parameters serving as the primary inferential target. Leveraging restricted maximum likelihood estimation and computationally efficient strategies, our approach overcomes statistical and computational challenges posed by high-dimensional spatiotemporal fMRI data. Validation on simulated data and empirical datasets demonstrates: (1) substantially reduced false-positive rates in postmortem rat fMRI; and (2) improved test–retest reliability and more robust standard errors for individual-level FC estimates compared to conventional correlation-based approaches in the Human Connectome Project (HCP) dataset. This work pioneers the systematic integration of LMMs into FC modeling, establishing a new paradigm for unbiased, reproducible connectomic analysis.
📝 Abstract
Resting-state brain functional connectivity quantifies the synchrony between activity patterns of different brain regions. In functional magnetic resonance imaging (fMRI), each region comprises a set of spatially contiguous voxels at which blood-oxygen-level-dependent signals are acquired. The ubiquitous Correlation of Averages (CA) estimator, and other similar metrics, are computed from spatially aggregated signals within each region, and remain the quantifications of inter-regional connectivity most used by neuroscientists despite their bias that stems from intra-regional correlation and measurement error. We leverage the framework of linear mixed-effects models to isolate different sources of variability in the voxel-level signals, including both inter-regional and intra-regional correlation and measurement error. A novel computational pipeline, focused on subject-level inter-regional correlation parameters of interest, is developed to address the challenges of applying maximum (or restricted maximum) likelihood estimation to such structured, high-dimensional spatiotemporal data. Simulation results demonstrate the reliability of correlation estimates and their large sample standard error approximations, and their superiority relative to CA. The proposed method is applied to two public fMRI data sets. First, we analyze scans of a dead rat to assess false positive performance when connectivity is absent. Second, individual human brain networks are constructed for subjects from a Human Connectome Project test-retest database. Concordance between inter-regional correlation estimates for test-retest scans of the same subject are shown to be higher for the proposed method relative to CA.
