Multi-domain feature analysis method of MI-EEG signal based on Sparse Regularity Tensor-Train decomposition.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Tensor analysis is valuable for multidomain characteristics in electroencephalography (EEG) studies.
• Existing EEG tensor methods face challenges with high dimensionality, hindering feature extraction.
• Traditional Tucker and Canonical Polyadic (CP) decomposition exhibit low computational efficiency and limited feature extraction capabilities.

Purpose of the Study:

• To address limitations of existing EEG tensor analysis methods.
• To propose an efficient and accurate tensor decomposition algorithm for EEG data.
• To enhance feature extraction and classification performance in EEG analysis.

Main Methods:

• Utilized Tensor-Train (TT) decomposition for EEG tensor analysis.
• Introduced a sparse regularization term to TT decomposition, creating the Sparse Regularized TT (SR-TT) algorithm.
• Validated the SR-TT algorithm using datasets from BCI competition III and BCI competition IV.

Main Results:

• The SR-TT algorithm achieved high classification accuracies of 86.38% (BCI III) and 85.36% (BCI IV).
• Demonstrated significant improvements in computational efficiency: 16.49-31.08 times faster than Tucker/CP on BCI III and 20.72-29.45 times faster on BCI IV.
• Enabled spatial feature extraction through tensor decomposition, visualized via brain topography.

Conclusions:

• The proposed SR-TT algorithm offers superior accuracy and generalization ability compared to state-of-the-art methods.
• SR-TT provides a substantial increase in computational efficiency for EEG tensor analysis.
• This novel approach offers new insights into tensor-based EEG analysis and brain activity visualization.