A self-organized recurrent neural network for estimating the effective connectivity and its application to EEG data

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Effective connectivity analysis is crucial for understanding brain region interactions.
• Dynamic systems in neuroscience benefit from dynamic analytical tools for improved results.

Purpose of the Study:

• To propose a novel dynamic approach for estimating effective connectivity.
• To introduce the Recurrent Neural Network - Neuron Growth Using Error Whiteness - Granger Causality (RNN-NGUEW-GC) method.

Main Methods:

• Utilizing Recurrent Neural Networks (RNNs) for time series and multivariate signal prediction.
• Employing Neuron Growth Using Error Whiteness (NGUEW) to determine optimal time lag via an error whiteness criterion.
• Integrating Granger causality for linear and nonlinear model analysis and defining an 'intensity of causality' indicator.

Main Results:

• The RNN-NGUEW-GC method demonstrated high accuracy on simulation data.
• Successfully predicted epileptic seizures using an EEG dataset.
• The method yields a self-organized network structure without requiring physiological information.

Conclusions:

• RNN-NGUEW-GC offers a superior approach to effective connectivity estimation compared to existing methods.
• The NGUEW technique for calculating time lag surpasses traditional multivariate auto-regressive models.
• The method's ability to create self-organized networks and its independence from physiological data are key advantages.