Using Robust Principal Component Analysis to Reduce EEG Intra-Trial Variability.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Practical brain-computer interfaces (BCIs) face challenges with signal variability within and across sessions, hindering reliable performance.
• Signal variability is a significant obstacle for the real-world application of BCI technology.

Purpose of the Study:

• To introduce Robust Principal Component Analysis (RPCA) as a method to address intra-trial signal variability in BCIs.
• To evaluate the effectiveness of RPCA in improving classification accuracy for BCI applications.

Main Methods:

• Applied Robust Principal Component Analysis (RPCA) as a preprocessing step to reduce trial variability in BCI data.
• Tested the RPCA approach on a workload detection task involving simulated drone piloting.
• Assessed the impact of RPCA on classification accuracy at both group and individual subject levels.

Main Results:

• RPCA significantly increased classification performance in the drone piloting task.
• Simulating RPCA online led to an average increase in class-balanced accuracy from 63.9% to 70.6% (p < 0.001).
• The improvement in accuracy was observed at both group and subject-level analyses.

Conclusions:

• Robust Principal Component Analysis (RPCA) is a viable and effective method for mitigating signal variability in brain-computer interfaces.
• The application of RPCA enhances the accuracy and reliability of BCIs, paving the way for more practical implementations.
• This study demonstrates the potential of RPCA to improve BCI performance in real-world scenarios.