ANFIS Granger Causality for Estimating Effective Brain Connectivity and Mutual Information for Connectivity Type Determination Using MEG and EEG Data: Application to Epilepsy

Document Type : Research Article

Authors

1 Master Graduate, School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran

2 Professor, School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran

Abstract

In the scientific community, it is well established that the brain is linked to neural disorders such as epilepsy, Alzheimer's, and depression, all of which can affect neural connectivity. These conditions can disrupt communication between different brain regions. To assess these changes, neuroscientists measure neural signals like EEG and MEG and analyze brain connectivity through scalp recordings. Various methods have been developed to evaluate intra-brain connectivity, including classical techniques such as Granger causality (GC), Mutual Information (MI), Directed Transfer Function (DTF), and Dynamic Causal Modeling (DCM). Recently, there has been increasing interest in applying neural networks as a modern approach across various fields. However, many existing methods suffer from low precision. This paper proposes the Adaptive Neuro-Fuzzy Inference System Granger Causality (ANFISGC) as a solution for measuring effective connectivity using EEG and MEG data. Our approach integrates symplectic geometry, ANFIS regression, and Granger causality, allowing for the detection of both linear and nonlinear causal information flow. This multivariate method can also differentiate between direct and indirect connectivity, enhancing its significance. Additionally, we utilized Mutual Information (MI) to evaluate the relationship between two variables, offering insights into the linearity or nonlinearity of connectivity. This measurement provides a further understanding of brain functionality. To assess the effectiveness of our approach, we conducted tests using simulated data and data from five epilepsy patients. The results show that measurements based on MEG data align well with clinical findings, while incorporating EEG data alongside MEG (in a multimodal approach) does not improve the results.

Keywords

Main Subjects

AUT Journal of Electrical Engineering
Volume 58, Issue 1
January 2026
Pages 31-44

Files

Share

How to cite

Statistics