Hand Motor Imagery Classification Using Effective Connectivity and Hierarchical Machine Learning in EEG Signals

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI) Brain Computer Interfaces (BCIs) typically use single Electroencephalogram (EEG) channels.
• Integrating inter-channel EEG relationships through effective brain connectivity analysis offers richer features.
• This study addresses the need for advanced feature extraction in MI-BCI systems.

Purpose of the Study:

• To identify robust and discriminative effective connectivity features from EEG signals.
• To develop a hierarchical machine learning framework for classifying left and right hand MI tasks.
• To enhance MI-BCI performance by leveraging brain connectivity patterns.

Main Methods:

• Effective connectivity was estimated using Granger Causality (GC) methods: Generalized Partial Directed Coherence (GPDC), Directed Transfer Function (DTF), and direct Directed Transfer Function (dDTF).
• Feature selection involved Kruskal-Wallis test and minimal-redundancy-maximal-relevance (mRMR) to identify significant causal connections.
• Support Vector Machine (SVM) was employed for final classification.

Main Results:

• The hierarchical BCI system achieved a maximum classification accuracy of approximately 84% across 29 subjects.
• Optimal performance was observed in the Mu (8-12 Hz) - Beta1 (12-15 Hz) frequency band using the GPDC method.
• The study demonstrated the effectiveness of GC-based connectivity features for MI discrimination.

Conclusions:

• A novel hierarchical automated BCI system was developed for effective MI task discrimination.
• The system successfully utilizes EEG effective connectivity features for classifying left and right hand motor imagery.
• This approach offers a promising advancement for developing more sophisticated BCI applications.