EEG classification of physiological conditions in 2D/3D environments using neural network

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) require high classification accuracy for reliable applications.
• Feature selection is crucial for enhancing BCI performance.
• Existing methods may not fully capture the complexity of physiological signals.

Purpose of the Study:

• To propose a new classification scheme for BCI applications.
• To evaluate the effectiveness of six nonlinear features for physiological state classification.
• To improve the accuracy of distinguishing between Eyes Open, Eyes Closed, and various game-playing conditions.

Main Methods:

• A novel scheme utilizing six nonlinear features: Sample entropy (SampEn), Composite permutation entropy index (CPEI), Approximate entropy (ApEn), Fractal dimension (FD), Hurst exponent (H), and Hjorth parameters (complexity and mobility).
• These features were employed as decision variables for classifying distinct physiological conditions.
• The scheme was tested on datasets representing Eyes Open (EO), Eyes Closed (EC), Game Playing 2D (GP2D), Game playing 3D active (GP3DA), and Game playing 3D passive (GP3DP).

Main Results:

• The proposed scheme successfully classified the physiological conditions.
• An overall classification accuracy of 88.9% was achieved.
• The selected nonlinear features demonstrated significant discriminative power.

Conclusions:

• The novel classification scheme effectively enhances BCI accuracy.
• Nonlinear features provide valuable information for discriminating between different physiological states.
• This approach holds promise for improving BCI system performance in various applications.