Personalized whole-brain Ising models with heterogeneous nodes capture differences among brain regions.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Ising models are used as simplified neural mass models to study complex brain dynamics from connectivity.
• Limitations in parameter estimation hinder their application to individual, high-resolution human neuroimaging data.
• Existing models often overlook node heterogeneity, despite regional differences in brain structure and dynamics.

Purpose of the Study:

• To develop an improved approach for fitting Ising models to functional MRI (fMRI) data for individual human brains.
• To investigate the impact of data binarization thresholds on model fitting and the representation of node heterogeneity.
• To enable personalized, biophysically interpretable modeling of structure-function mapping in the brain.

Main Methods:

• Developed an enhanced method for fitting Ising models to 360-region fMRI data.
• Incorporated derivation of an initial guess model from group data, optimization of simulation temperature, and two stages of Boltzmann learning (group and individual data).
• Utilized GPU acceleration for computational efficiency and analyzed effects of data binarization thresholds and external fields.

Main Results:

• Higher fMRI data binarization thresholds decrease correlation with functional connectivity but increase node external field heterogeneity and correlations with structural MRI features (myelination, cortical folding).
• A specific threshold choice balances goodness-of-fit with regional heterogeneity, yielding a more realistic brain network model.
• The approach allows for personalized modeling of structure-function relationships across the whole brain.

Conclusions:

• The developed method enhances the fitting of Ising models to individual neuroimaging data, accounting for node heterogeneity.
• This approach bridges connectomics and translational research by enabling personalized, biophysically interpretable brain network modeling.
• It aids in understanding individual differences in brain network organization and structure-function mapping.