A hybrid framework (neural mass model + ML) for SC-to-FC prediction

🌈 A hybrid framework (neural mass model + ML) for SC-to-FC prediction

The current workflow simulates brain functional connectivity (FC) from structural connectivity (SC) with a neural mass model. Gradient descent is applied to optimize the parameters in the neural mass model.

The pipeline contains the following components:

• Neural Mass Model (models/torch_neural_mass.py): It is an ODE system that describes the neural activities over time. The Wilson-Cowan model is implemented here with a connected network setting - each neural region is considered as a node in the brain network and connected via SC. The Wilson-Cowan model assumes each node contains two types of neural populations: the excitatory and inhibitory cells. The definition can be found here and here.
• Hemodynamic Model (models/hrf_torch.py): This module down samples and transforms the neural activities into Blood Oxygen Level Dependence (BOLD) signals. The code is adapted from the Virtual Brain implementation of the Balloon model.

Requirement

PyTorch (my version is 1.10.0)

Usage

An example of running the pipeline can be found at run.sh. Please update path to your data.