A Comparative Evaluation of Molecular Connectivity and Covariance Approaches

Abstract

Advances in high-temporal-resolution functional positron emission tomography (fPET) now enable the assessment of metabolic associations between brain regions, providing a molecular complement to functional connectivity derived from fMRI. However, the distinction between molecular connectivity (MC) and covariance (mCov) remains conceptually and methodologically inconsistent across studies. This work systematically compares major analytical approaches for MC and mCov to clarify their assumptions, dependencies, and interpretational boundaries with the most used radiotracer [18F]fluorodeoxyglucose to obtain metabolic connectivity (M-MC). Twenty healthy participants underwent ultra-high-sensitivity fPET acquisitions on a large axial field-of-view scanner. M-MC was estimated using CompCor, spatio-temporal filtering, third-order polynomial detrending, baseline normalization and Euclidean distance at multiple temporal resolutions. mCov was assessed from SUVR images with subsequent network- or subject-specific matrix computation using independent component analysis, principal component analysis, or jackknife perturbation. Results demonstrate that while all MC methods are valid, CompCor and Euclidean distance perform optimally at high temporal resolutions (1-16s), whereas polynomial and spatio-temporal filters are more robust at lower sampling rates (>16s). mCov offers a population-level characterization of metabolic organization with the option to derive relative-to-group single-subject maps. This comparison provides methodological clarity and supports standardized use of molecular network analyses now integrated into the open-source fPET toolbox.