Estimating Effective Connectivity by Recurrent Generative Adversarial Networks.

Area of Science:

• Neuroinformatics
• Brain Informatics
• Computational Neuroscience

Background:

• Estimating effective connectivity from functional magnetic resonance imaging (fMRI) time series data is crucial in neuroinformatics.
• Current methods struggle with accuracy due to high noise and small sample sizes inherent in fMRI data.

Purpose of the Study:

• To propose a novel framework, EC-RGAN, for accurate effective connectivity estimation from fMRI data.
• To address limitations of existing methods in handling noisy and limited fMRI datasets.

Main Methods:

• Developed a framework based on recurrent generative adversarial networks (RGANs).
• The generator utilizes recurrent neural networks to simulate fMRI time series for brain regions.
• The discriminator distinguishes between real and generated fMRI data distributions.

Main Results:

• EC-RGAN successfully estimates effective connectivity by leveraging causal parameters from trained generators.
• Experimental results on simulated and real fMRI data confirm the framework's efficacy.
• The model demonstrates improved accuracy in effective connectivity estimation.

Conclusions:

• EC-RGAN offers a robust and effective approach for estimating brain connectivity from fMRI data.
• The proposed framework shows promise for advancing neuroinformatics and brain informatics research.
• This method has the potential to overcome common challenges in analyzing fMRI time series data.