Are linear models’ apparent superiority due to neuroimaging limitations?

Ascertain whether the observed superiority of linear models in describing macroscopic resting-state brain dynamics arises from intrinsic linearity at the measured scale or instead from limitations of current neuroimaging modalities (e.g., temporal resolution, noise characteristics, preprocessing), thereby clarifying the appropriate level of model complexity for large-scale brain activity.

Background

Recent work indicates that linear models may outperform nonlinear models for macroscopic resting-state dynamics, despite the intrinsic nonlinearity of neuronal processes. This raises a key question about whether the finding reflects true properties of large-scale dynamics or is an artifact of measurement constraints.

Resolving this will inform when linear approximations are sufficient and when richer nonlinear models are necessary, guiding both theoretical development and practical model fitting for fMRI/MEG/EEG data.

References

However, recent work has indicated that linear models may outperform nonlinear ones in the description of macroscopic resting-state dynamics. Whether this is a natural limitation of current neuroimaging techniques is yet to be determined.

— Nonequilibrium physics of brain dynamics  (2504.12188 - Nartallo-Kaluarachchi et al., 16 Apr 2025) in Subsubsection 'The linear Langevin model' within 'Model-based analysis of neural activity' (Section: Analysis of neuroimaging and continuous-valued recordings)

This discrepancy leaves a critical ambiguity: it remains unclear whether unpaired analyses are simply confounded by high inter-individual variance, or if identifying the left-handed neural equivalent of a right-hander's dominant grasp requires more sophisticated metrics sensitive to the directionality of information.

— Cortical information transfer reveals conserved hemispherical network dynamics across human handedness  (2609.10870 - Ramos et al., 9 Sep 2026) in Section Introduction