Classification of motor imagery EEG with ensemble RNCA model

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI) based Brain-Computer Interface (BCI) systems aid individuals with motor function impairments.
• Redundant electroencephalography (EEG) channels often degrade MI task performance.
• Accurate identification of neural regions is crucial for effective BCI operation.

Purpose of the Study:

• To develop a novel method for precise identification of neural regions involved in motor movements for MI-BCI.
• To enhance the classification accuracy and reduce computational complexity in MI-BCI systems.
• To improve the performance of BCI systems for individuals with neurophysiological impairments.

Main Methods:

• Utilized an ensemble regulated neighborhood component analysis (ERNCA) for neural region identification.
• Employed statistical, frequency, spatial, and transform-based features to minimize misclassification.
• Applied gradient boosting for relevant feature selection and Bayesian optimization for ensemble classifier tuning.

Main Results:

• Achieved high classification accuracies of 97.22% for Dataset IIIa and 91.62% for Dataset IVa.
• Demonstrated strong real-time performance with an accuracy of 93.75%.
• Outperformed four benchmark methods, showing significant accuracy improvements across datasets.

Conclusions:

• The ERNCA method effectively identifies motor-related neural regions, improving MI-BCI performance.
• Feature selection and optimized classification significantly enhance system accuracy and efficiency.
• The study highlights the concentration of motor functions in the frontal and central cortex regions.