Motor Imagery Classification Based on EEG Sensing with Visual and Vibrotactile Guidance.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Motor imagery (MI) is a key component in brain-computer interfaces (BCIs).
• Electroencephalography (EEG) is commonly used to record brain activity for MI-based BCIs.
• Evaluating different machine learning classifiers is crucial for optimizing BCI performance.

Purpose of the Study:

• To compare the performance of six different classifiers for EEG-based motor imagery detection.
• To investigate the impact of various guidance methods (visual, vibrotactile) on classifier accuracy.
• To assess the effect of data preprocessing techniques on classification outcomes.

Main Methods:

• EEG datasets of motor imagery tasks were analyzed.
• Six classifiers were evaluated: LDA, SVM, RF, and three CNN variants (ResNet-based).
• Data preprocessing involved filtering and feature extraction, with varying passbands investigated.

Main Results:

• A ResNet-based convolutional neural network (CNN) significantly outperformed other classifiers.
• Vibrotactile guidance notably improved classification accuracy, especially for simpler models.
• Preprocessing using low-frequency signal features enhanced classification performance.

Conclusions:

• ResNet-based CNNs are highly effective for EEG-based motor imagery detection.
• Vibrotactile guidance offers a significant advantage for improving BCI accuracy.
• Optimized data preprocessing is essential for robust EEG-based BCI systems.