EEG multi-domain feature transfer based on sparse regularized Tucker decomposition.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) tensor analysis reveals brain activity and interactions.
• Subject variability in EEG data challenges existing tensor decomposition methods, leading to non-objective classification.
• Traditional Tucker decomposition faces issues with feature dimension explosion.

Purpose of the Study:

• To propose a novel EEG tensor transfer algorithm, Tensor Subspace Learning based on Sparse Regularized Tucker Decomposition (TSL-SRT).
• To address the non-objectivity and feature dimension explosion issues in EEG tensor analysis.
• To ensure features across subjects are distributed in the same domain for improved analysis.

Main Methods:

• Developed TSL-SRT algorithm integrating feature transfer and sparse regularized Tucker decomposition.
• Treated new EEG samples as the target domain and original samples as the source domain for feature projection.
• Employed a redundant EEG feature screening algorithm to mitigate dimension explosion.

Main Results:

• Achieved classification accuracies of 77.8%, 73.2%, and 75.3% on three Brain-Computer Interface (BCI) datasets.
• Visualizations confirmed TSL-SRT's effectiveness in extracting active brain region features for BCI tasks.
• Demonstrated simultaneous extraction of multi-domain features from different subjects within a unified domain.

Conclusions:

• TSL-SRT offers a novel and effective method for EEG tensor analysis, overcoming limitations of existing algorithms.
• The algorithm ensures objectivity and handles feature dimension issues in cross-subject EEG data.
• TSL-SRT facilitates a unified domain for analyzing diverse subject features in BCI applications.