Systematic evaluation of structural connectome thresholding in whole-brain network modelling

Abstract

Large-scale whole-brain network models rely on structural connectomes to constrain simulated neural dynamics, yet the optimal processing of these anatomical scaffolds is not fully established. Here, we investigate the impact of structural connectome thresholding on whole-brain model performance to advance precision individual modelling. We measure model goodness-of-fit to both static functional connectivity and dynamic functional connectivity across a comprehensive spectrum of network densities, spatial parcellations, and model complexities using neuroimaging datasets of healthy individuals and a clinical cohort of post-stroke patients. We demonstrate that the optimal structural sparsity is highly resolution dependent. In coarse- grained parcellations, proportional thresholding enhances the model’s fit to static functional connectivity but relies on denser connectome to maintain dynamic functional fits. Conversely, fine-grained models require stringent connectome thresholding which simultaneously optimizes both static and dynamic functional fits. Taken together, these results indicate that the uncritical use of raw structural connectomes introduces suboptimal dynamical regimes, establishing resolution-tailored thresholding as an indispensable step for constructing precision brain network models.