Long-term decoding of arm movement using Spatial Distribution of Neural Patterns.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Daily variations in subject motivation and behavior challenge the reliability of Brain Computer Interfaces (BCIs) that use local field potentials (LFPs).
• Standard pattern recognition algorithms struggle with non-stationary data, leading to performance degradation over time.
• Models trained on one day often fail to accurately decode user intent on subsequent days.

Purpose of the Study:

• To develop a novel algorithm for adaptive spatial features to address variability in LFP signals.
• To improve the long-term decoding accuracy and practical usability of BCIs.
• To enable robust BCI performance despite day-to-day changes in user state.

Main Methods:

• Proposed an algorithm to capture local spatial variability within LFP patterns.
• Implemented adaptive spatial features to create more robust signal representations.
• Tested the algorithm's performance on decoding eight movement directions over an extended period.

Main Results:

• Achieved over 95% accuracy in decoding eight distinct movement directions.
• Demonstrated successful decoding two weeks after the initial training session.
• The adaptive spatial features effectively mitigated the impact of daily signal non-stationarity.

Conclusions:

• Adaptive spatial features offer a promising solution for enhancing the long-term stability and accuracy of BCIs.
• The proposed algorithm significantly improves decoding performance in practical BCI applications.
• This approach addresses a key limitation in current BCI technology, paving the way for more reliable human-computer interaction.