[Tensor Feature Extraction Using Multi-linear Principal Component Analysis for Brain Computer Interface].

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) enable device control via electroencephalogram (EEG) signals.
• Traditional methods like PCA and 2DPCA face limitations in processing multichannel EEG data in tensor form.
• Efficient feature extraction and dimensionality reduction are crucial for BCI performance.

Purpose of the Study:

• To develop and evaluate a Multi-linear Principal Component Analysis (MPCA) framework for EEG signal processing in BCIs.
• To address the limitations of existing dimensionality reduction techniques for tensor-formatted EEG data.
• To enhance feature extraction and classification accuracy for BCI applications.

Main Methods:

• A novel MPCA framework was developed for processing multichannel EEG signals represented as tensors.
• The method utilizes tensor-matrix projection for dimensionality reduction and feature extraction.
• Fisher linear classifier was employed for feature classification, with extensive parameter tuning (P and Q).

Main Results:

• The MPCA framework was tested on BCI competition datasets (II dataset 4, N dataset 3).
• Both second-order (time-space) and third-order (time-space-frequency) EEG tensor representations were analyzed.
• Achieved highest accuracy rates of 81.0% and 40.1% for second-order tensors, and 76.0% and 43.5% for third-order tensors, outperforming other methods.

Conclusions:

• The MPCA framework offers a superior approach for EEG signal processing in BCIs compared to traditional methods.
• MPCA effectively reduces dimensionality and extracts relevant features from complex EEG data.
• This method holds promise for improving the accuracy and performance of EEG-based BCI systems.