A Novel Feature Extraction Framework for Four Class Motor Imagery Classification using Log Determinant Regularized...

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-Computer Interface (BCI) systems facilitate communication for individuals by leveraging Electroencephalography (EEG) signals.
• Motor Imagery (MI) based BCIs are crucial for enabling interaction with the external environment.
• Accurate classification of MI tasks is essential for effective BCI operation.

Purpose of the Study:

• To propose a novel and robust feature extraction and classification framework for four-class Motor Imagery (MI) classification.
• To enhance classification accuracy in BCI systems by improving feature robustness against noise and outliers.
• To introduce a new architecture combining log-determinant (log-det) based Regularized Riemannian mean (LDRRM) with linear Support Vector Machine (SVM).

Main Methods:

• Development of a novel framework utilizing log-determinant (log-det) based Regularized Riemannian mean (LDRRM) for feature extraction.
• Integration of LDRRM with a linear Support Vector Machine (SVM) classifier for a four-class MI classification task.
• Evaluation of the proposed framework on a publicly available four-class MI dataset (BCI Competition IV, dataset 2a).

Main Results:

• The proposed LDRRM classification framework achieved a mean classification accuracy of 69.12% on the benchmark dataset.
• The LDRRM approach demonstrated improved robustness of extracted features against noise and outliers.
• The framework outperformed existing studies by achieving 1.54% higher classification accuracy.

Conclusions:

• The proposed LDRRM framework offers a robust and effective method for feature extraction and classification in four-class MI BCI systems.
• This approach significantly enhances classification accuracy, contributing to more reliable BCI applications.
• The LDRRM method shows promise for improving human-computer interaction through advanced BCI technology.