Hemispheric-Specific Coupling Improves Modeling of Functional Connectivity Using Wilson–Cowan Dynamics
Ramiro Plüss, Hernán Villota, Patricio Orio · 2025
Read Hemispheric-Specific Coupling Improves Modeling of Functional Connectivity Using Wilson–Cowan Dynamics by Ramiro Plüss, Hernán Villota, Patricio Orio (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-22951.txt.
Introduction
Ramiro Pluss, Hernan Villota, and Patricio Orio refine whole-brain simulations by treating the two hemispheres differently in a 2025 paper. Neural mass models simulate resting brain activity from the wiring map between regions. The classic Wilson-Cowan setup they extend uses one coupling strength style throughout. The authors split coupling into two kinds: links inside a hemisphere and links crossing between them.
Run on real cortical wiring and resting scan data from healthy controls and people with schizophrenia, the split model produced simulated activity patterns that matched the real functional connectivity more closely. The gain, they argue, comes from respecting anatomy: the brain's halves are wired to each other more sparsely and differently than they are wired internally. Hemisphere-aware coupling, they suggest, should be standard in large-scale brain models.
Catalog record
| CREATOR | Ramiro Plüss, Hernán Villota, Patricio Orio |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-22951
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hemispheric-Specific Coupling Improves Modeling of Functional Connectivity Using Wilson–Cowan Dynamics Ramiro Plüss, Hernán Villota, Patricio Orio Abstract Large-scale neural mass models have been widely used to simulate resting-state brain activity from structural connectivity. In this work, we extend a well-established Wilson–Cowan framework by introducing a novel hemispheric-specific coupling scheme that differentiates between intra-hemispheric and inter-hemispheric structural interactions. We apply this model to empirical cortical connectomes and resting-state fMRI data from matched control and schizophrenia groups. Simulated functional connectivity is computed from the band-limited envelope correlations of regional excitatory activity and compared against empirical functional connectivity matrices. Our results show that incorporating hemispheric asymmetries enhances the correlation between simulated and empirical functional connectivity, highlighting the importance of anatomically-informed coupling strategies in improving the biological realism of large-scale brain network models. Keywords: Wilson–Cowan model, structural connectivity, functional connectivity, schizophrenia, neural dynamics, hemispheric coupling 1. Introduction Schizophrenia is a severe neuropsychiatric disorder characterized by profound disruptions in thought processes, perception, and affect regulation. Mounting evidence indicates that these symptoms are associated with altered functional connectivity (FC) across large-scale brain networks, particularly within and between the default mode, frontoparietal control, and salience networks [ 1 , 2 ] . Understanding how these functional abnormalities arise from the underlying anatomical organization, or structural connectivity (SC), remains a central question in computational psychiatry. Structural connectomes derived from diffusion MRI provide a static scaffold upon which neural dynamics unfold. However, the mapping between SC and FC is complex due to nonlinear interactions and modulatory processes that shape brain activity. Computational models of whole-brain dynamics offer a principled framework to investigate this SC–FC relationship. In particular, the Wilson–Cowan neural mass model provides a biologically grounded formalism to simulate excitatory and inhibitory population dynamics, enabling the generation of realistic large-scale activity patterns [ 3 , 4 ] . In this work, we simulate FC from empirical cortical SC matrices obtained from matched control and schizophrenia groups. To isolate cortical dynamics and minimize confounds related to deep nuclei, we restrict our analysis to cortical regions. We first implement a classic whole-brain model with uniform global coupling strength ( G ), and then introduce a novel hemispheric-specific coupling strategy that distinguishes intra- ( G_{1} ) and inter-hemispheric ( G_{2} ) connectivity. This extension is motivated by anatomical and functional asymmetries previously reported in both healthy and clinical populations [ 1 ] . We evaluate the capacity of each model variant to reproduce empirical resting-state FC, using Pearson correlation as a measure of SC–FC correspondence. Additionally, we analyze group differences and explore the role of hemispheric specificity in capturing altered integration patterns observed in schizophrenia. Our results show that introducing hemispheric coupling parameters improves the model’s performance and may help reveal biologically meaningful asymmetries in large-scale brain dynamics. 2. Material and Methods 2.1. Empirical Dataset and Brain Network Estimation We employed a publicly available dataset containing diffusion and functional MRI scans from 27 patients with schizophrenia and 27 demographically matched healthy controls [ 2 , 5 ] . These data were acquired using a 3T Siemens Trio scanner and include high-resolution T1-weighted images, diffusion spectrum imaging (DSI), and resting-state fMRI. Participants in the patient group met DSM-IV criteria for schizophrenia or schizoaffective disorder, and were recruited from the Lausanne University Hospital. Brain network construction was performed using the Connectome Mapping Toolkit [ 6 ] . Cortical segmentation was based on the Lausanne 2008 multi-scale parcellation scheme [ 7 ] , from which we selected the 83-region resolution to balance anatomical specificity with computational feasibility. SC matrices were derived from deterministic streamline tractography applied to the DSI data, with edge weights defined as normalized streamline counts accounting for surface area and length biases [ 8 ] . FC was estimated as the absolute Pearson correlation between regional mean BOLD signals after standard preprocessing, including motion correction, nuisance regression, spatial smoothing, and band-pass filtering [ 9 , 10 ] . This framework provides subject-level structural and functional networks with matched parcellations, enabling direct comparison and simulation of brain dynamics constrained by individual connectomes. To reduce inter-subject variability and highlight group-level trends, we computed group-average SC and FC matrices separately for the control (CTRL) and schizophrenia (SCHZ) cohorts. Since all individual SC and FC matrices are symmetric by construction—owing to the undirected nature of tractography and the use of Pearson correlation—the resulting group-average matrices are also symmetric. This approach allows us to focus on global network properties and systematic differences between groups, rather than individual-specific fluctuations. By simulating neural dynamics on the group-averaged SC using the Wilson–Cowan model, we can directly assess how alterations in anatomical connectivity influence the emergence of functional interactions at the population level. Furthermore, the use of averaged matrices facilitates a controlled exploration of model parameters, such as global versus hemispheric-specific coupling schemes, enabling us to investigate how different modes of interaction (intra-hemispheric vs. inter-hemispheric connectivity) affect the model’s ability to reproduce empirical FC. This is particularly relevant in schizophrenia, where disruptions in inter-hemispheric integration have been consistently reported. Finally, from a translational perspective, the group-level SC and FC matrices serve as clinical reference profiles, against which deviations at the individual level could be later evaluated. This provides a meaningful scaffold for assessing the impact of pathological alterations and testing hypotheses about network-level mechanisms underlying psychiatric disorders. The 83-region resolution includes both cortical and subcortical structures. To restrict the analysis to cortical dynamics, we excluded subcortical regions and the brainstem, resulting in 68 cortical ROIs (34 per hemisphere). This reduction ensures compatibility with the Wilson–Cowan neural mass model, which is defined at the mesoscopic cortical level. All modeling and simulation steps were performed on these 68\times 68 group-average matrices. 2.2. Wilson–Cowan Network Model We employed a whole-brain neural mass model based on the Wilson–Cowan equations [ 3 , 4 ] , which describe the dynamics of coupled excitatory ( E ) and inhibitory ( I ) neural populations. Each cortical region k is modeled as a pair of interacting populations governed by Eq. ( 1 ) and Eq. ( 2 ). \displaystyle\tau_{e}\frac{dE_{k}}{dt} \displaystyle=-E_{k}+(1-r_{e}E_{k})S_{e}\left(a_{ee}E_{k}-a_{ei,k}I_{k}+G\sum_{\begin{subarray}{c}l=1\\ l\neq k\end{subarray}}^{N}C_{kl}E_{l}+P\right) (1) \displaystyle\tau_{i}\frac{dI_{k}}{dt} \displaystyle=-I_{k}+(1-r_{i}I_{k})S_{i}\left(a_{ie}E_{k}-a_{ii}I_{k}+Q\right) (2) Here, \tau_{e} and \tau_{i} are the time constants for excitatory and inhibitory populations, respectively; r_{e} and r_{i} modulate population refractoriness; and a_{\alpha\beta} are synaptic weight parameters. The matrix C_{kl} encodes the normalized structural connectivity between regions, and G is a global coupling parameter. P and Q represent constant external inputs. The transfer function S(x;\mu,\sigma) is a sigmoidal nonlinearity that governs the population response, where \mu defines the activation threshold and \sigma controls the slope of the transition. Its general form is given by Eq. ( 3 ). S(x)=\frac{1}{1+\exp\left(-\frac{x-\mu}{\sigma}\right)} (3) This formulation has been widely used to simulate large-scale brain dynamics and to investigate the emergence of functional connectivity from anatomical structure [ 11 ] . To derive simulated FC from the model, we extract the excitatory activity E_{k}(t) from each region after a transient period and compute its analytic signal using the Hilbert transform. The magnitude of this signal provides the amplitude envelope, which captures the slow oscillatory modulation of neural activity. Simulated FC is then defined as the Pearson correlation between the amplitude envelopes of all pairs of regions. This method follows standard procedures for extracting resting-state FC from band-limited neural