Classification of BCI Multiclass Motor Imagery Task Based on Artificial Neural Network

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Electroencephalography (EEG)-based motor imagery (MI) signals are crucial for brain-computer interfaces (BCI).
• Extracting user-specific features and enhancing classifier accuracy remain challenges in MI-BCI.
• Current BCI systems require improved performance for individuals with motor disabilities.

Purpose of the Study:

• To propose and evaluate a novel artificial neural network (ANN) architecture for enhanced motor imagery classification in BCI.
• To compare the performance of the proposed ANN against traditional classification algorithms.
• To investigate the impact of Batch Normalization layers on ANN learning time and accuracy.

Main Methods:

• Utilized EEG data from the BCI Competition IV-2a dataset.
• Implemented and compared four classification algorithms: Quadratic Discriminant Analysis, k-Nearest Neighbors, Linear Discriminant Analysis, and a proposed ANN architecture.
• Employed feature extraction techniques including time domain parameters, band power, signal power, and wavelet packet decomposition (WPD).
• Incorporated Batch Normalization layers into the ANN architecture to optimize learning and accuracy.

Main Results:

• The proposed ANN architecture achieved a kappa score of 0.5545 and 58.42% accuracy on the BCI Competition IV-2a dataset.
• The modified ANN, utilizing frequency features from WPD and spatial features from Common Spatial Pattern (CSP), demonstrated superior classification performance.
• Batch Normalization layers improved the learning time and accuracy of the neural network, while also providing regularization.

Conclusions:

• The developed ANN method offers a significant improvement for motor imagery classification in BCI applications.
• Combining WPD and CSP for feature extraction with the enhanced ANN architecture is effective for improving BCI performance.
• The proposed approach holds promise for advancing communication and control for individuals with motor impairments.