A Large-scale Neural Model Inversion Framework for Effective Connectivity Estimation

Area of Science:

• Computational neuroimaging
• Systems neuroscience
• Neuroinformatics

Background:

• Estimating large-scale brain-wide effective connectivity (EC) from resting-state functional MRI (rs-fMRI) is a significant challenge.
• Existing methods may struggle with the computational demands of whole-brain analysis.

Purpose of the Study:

• To develop and validate a novel computational framework, Large-scale nEural Model Inversion (LEMI), for efficient and accurate estimation of large-scale brain-wide EC.
• To assess the framework's performance using simulations and apply it to empirical data for neuroscientific insights.

Main Methods:

• Developed LEMI utilizing a linear neural mass model and a Kalman-filter based gradient descent algorithm.
• Validated LEMI's accuracy and efficiency in recovering model parameters for a 100-region network within 90 minutes using ground-truth simulations.
• Applied LEMI to an Alzheimer's Disease Neuroimaging Initiative (ADNI) rs-fMRI dataset.

Main Results:

• LEMI accurately and efficiently recovered model parameters in large-scale network simulations.
• The framework successfully estimated intra-regional and inter-regional connection strengths.
• Application to ADNI data revealed a widespread reduction in the excitation-inhibition (E-I) ratio in Alzheimer's disease patients.

Conclusions:

• LEMI offers an efficient and accurate computational framework for estimating large-scale effective connectivity from rs-fMRI data.
• The framework enables the exploration of brain-wide E-I balance, providing insights into neurological conditions like Alzheimer's disease.
• LEMI facilitates advancements in computational neuroimaging and understanding brain mechanisms.