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

Area of Science:

• Computational Neuroscience
• Neuroimaging Analysis
• Network Science

Background:

• Ising models are used to study brain dynamics but face limitations with individual neuroimaging data.
• Existing models often overlook node heterogeneity, despite regional differences in brain structure and function.
• Personalized modeling is crucial for understanding individual brain network organization.

Purpose of the Study:

• To develop and validate an improved approach for fitting Ising models to high-resolution functional MRI data for individual subjects.
• To investigate the impact of data binarization thresholds on model fitting and the representation of node heterogeneity.
• To explore the relationship between Ising model parameters and structural MRI features, such as myelination and cortical folding.

Main Methods:

• Developed a multi-stage fitting process including group data initialization, simulation temperature optimization, and dual-stage Boltzmann learning (group then individual data).
• Utilized GPU acceleration to manage the computational demands of the modeling approach.
• Analyzed the effects of fMRI data binarization thresholds on goodness-of-fit, external field heterogeneity, and correlations with structural MRI metrics.

Main Results:

• Higher fMRI binarization thresholds reduced functional connectivity correlation but increased node external field heterogeneity and its correlation with structural features.
• Identified an optimal binarization threshold balancing goodness-of-fit with intrinsic regional heterogeneity.
• Demonstrated that the refined Ising models better capture the brain's network of heterogeneous nodes.

Conclusions:

• The enhanced Ising model fitting approach enables personalized, biophysically interpretable modeling of whole-brain structure-function mapping.
• This method can advance the understanding of individual differences in brain network organization.
• It bridges the gap between connectomics and translational neuroscience research by integrating network and regional perspectives.