Neurophysiological predictors of deep learning based unilateral upper limb motor imagery classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery-based brain-computer interfaces (BCIs) decode movement intentions from imagined actions.
• Decoding unilateral upper limb motor imagery remains challenging despite deep learning advancements.
• Traditional machine learning is favored over deep learning for BCI analysis.

Purpose of the Study:

• To investigate the influence of neurophysiological features on classification accuracy in unilateral motor imagery BCIs.
• To compare deep learning models with traditional machine learning classifiers for motor imagery decoding.
• To explore the relationship between alpha and beta band power and BCI performance.

Main Methods:

• Classified imagined right elbow flexion/extension using EEG data from three deep learning models (EEGNet, FBCNet, NFEEG) and two machine learning classifiers (SVM, LDA).
• Analyzed absolute and relative alpha and beta power from frontal, fronto-central, and central electrodes during eyes-open and eyes-closed resting states.
• Correlated neurophysiological features with classifier accuracies to predict performance.

Main Results:

• Relative alpha band power negatively correlated with classifier accuracies.
• Absolute and relative beta band power positively correlated with classifier accuracies.
• Most ipsilateral EEG channels showed significant correlations with classifier accuracies, particularly for machine learning classifiers.

Conclusions:

• Neurophysiological features, specifically alpha and beta band power, can predict BCI classification accuracy for unilateral motor imagery.
• The findings suggest unique neurophysiological mechanisms, including ipsilateral inhibition and attentional processes, are involved in unilateral motor imagery.
• Deep learning models show potential but require further exploration in the context of motor imagery BCI analysis.