Overcoming Long-Term Variability in Local Field Potentials Using an Adaptive Decoder.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Long-term signal variability in intracortical recordings is a major challenge for brain-computer interfaces (BCI).
• Accurate and reliable estimation of subject behavior over extended periods requires overcoming neural signal time instability.

Purpose of the Study:

• To present a novel adaptive decoder for BCIs that addresses long-term neural signal variability.
• To characterize behavioral tasks using multiple neural patterns and adapt to variations by identifying stable neural patterns.

Main Methods:

• Developed a decoder using redundant sparse regression models to adapt to daily variations in neural patterns.
• Implemented unsupervised and semi-supervised learning frameworks for adaptation with minimal user feedback.
• Investigated decoder performance under varying external forces and over extended durations (up to 42 days).

Main Results:

• The adaptive decoder achieved 95% accuracy in predicting eight hand-movement directions over two weeks without external forces.
• Accuracy remained high at 85% in later sessions up to 42 days, even when monkeys countered external field forces.
• The decoder demonstrated effective operation with or without manual user intervention.

Conclusions:

• The proposed adaptive decoder effectively mitigates long-term variability in neural signals for BCIs.
• This approach enhances the robustness and reliability of neural decoders, particularly in challenging conditions with external forces.
• The decoder's ability to operate autonomously or with minimal feedback can significantly reduce user frustration and improve BCI system performance.