To Explore the Potentials of Independent Component Analysis in Brain-Computer Interface of Motor Imagery

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery (MI) brain-computer interfaces (BCI) are crucial for assistive technologies.
• Independent Component Analysis (ICA) offers potential for spatial filtering in MI-BCI.
• Evaluating classical ICA algorithms for MI-BCI performance is essential.

Purpose of the Study:

• To experimentally evaluate the potential of ICA in MI-BCI.
• To compare the performance of four classical ICA algorithms and a simplified Infomax (sInfomax) against Common Spatial Pattern (CSP).
• To introduce novel performance metrics for assessing ICA-MIBCI.

Main Methods:

• Developed an algorithmic framework for ICA-based MI-BCI (ICA-MIBCI).
• Evaluated Infomax, FastICA, Jade, Sobi, and sInfomax algorithms.
• Utilized self-test accuracy and number of invalid ICA filters as performance indexes.
• Compared ICA-MIBCI with CSP-MIBCI using experimental and online tests.

Main Results:

• sInfomax demonstrated superior session-to-session and subject-to-subject transferability compared to CSP.
• Classical ICA variants (FastICA, Jade, Sobi) showed poorer classification accuracy and stability than sInfomax and CSP.
• Online experiments confirmed the practicability of sInfomax-based MIBCI.

Conclusions:

• The simplified Infomax algorithm shows significant promise for improving ICA-based MI-BCI.
• Conventional ICA methods may overfit real-world EEG data, limiting their effectiveness.
• Further research into ICA variants could enhance practical MI-BCI implementations.