Overcoming measurement time variability in brain machine interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Local Field Potentials (LFP) are crucial for brain-computer interfaces (BCIs).
• Signal instability across recording sessions poses a significant challenge for BCI performance.
• Accurate decoding of movement intention from neural signals is vital for assistive technologies.

Purpose of the Study:

• To introduce a novel subspace learning approach for analyzing multi-channel LFP data.
• To enhance the robustness of neural decoding against inter-session signal variability.
• To improve the accuracy of movement direction decoding using LFP signals.

Main Methods:

• Subspace learning was applied to extract recurrent features from multi-channel LFP.
• The method was evaluated for decoding 8-direction movements.
• Performance was assessed by training on two sessions and testing on a third.
• The approach was combined with Error-Correcting Output Codes (ECOC) and Common Spatial Patterns (CSP) classifiers.

Main Results:

• The subspace learning method effectively addressed signal instability by capturing consistent neural patterns.
• Movement direction decoding performance showed significant improvement.
• Decoding Power (DP) increased from 76% to 88% for a subject with high inter-session variability.
• DP improved from 86% to 90% for a subject with lower variability.

Conclusions:

• Subspace learning offers a robust solution for decoding neural signals affected by inter-session variability.
• This method enhances the reliability and accuracy of BCIs utilizing LFP data.
• The findings suggest potential for improved neuroprosthetic control and assistive devices.