mpdcm: A toolbox for massively parallel dynamic causal modeling.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging

Background:

• Dynamic Causal Modeling (DCM) for fMRI is crucial for inferring effective brain connectivity.
• DCM utilizes biophysical models linking neuronal activity to BOLD signals.
• Current DCM simulations present significant computational challenges.

Purpose of the Study:

• Introduce Massively Parallel Dynamic Causal Modeling (mpdcm) to overcome computational bottlenecks in DCM.
• Provide a GPU-accelerated toolbox for efficient biophysical simulations.

Main Methods:

• mpdcm utilizes graphical processing units (GPUs) for parallel simulation generation.
• Employs a low storage explicit Runge-Kutta scheme optimized for GPU architecture.
• The mpdcm toolbox is available under the GPLv3 license.

Main Results:

• mpdcm achieves high-efficiency simulation generation without sacrificing accuracy.
• Demonstrates the utility of mpdcm for computationally intensive sampling algorithms like thermodynamic integration and parallel tempering.

Conclusions:

• mpdcm offers a two-order-of-magnitude efficiency improvement over standard SPM implementations.
• Facilitates advanced sampling methods, such as parallel tempering, by providing efficient, parallel model simulations.
• GPU accessibility makes advanced DCM analyses feasible without requiring large computing clusters or specialized expertise.