Inhibition of Long-Term Variability in Decoding Forelimb Trajectory Using Evolutionary Neural Networks With...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-machine interfaces (BMIs) face challenges with long-term decoding stability due to variations in neural recording conditions.
• Existing decoders may not be robust to these changes, especially with limited training data.

Purpose of the Study:

• To introduce an error feedback mechanism to neural decoders to address variability in neural recording conditions.
• To develop a robust neural decoder for stable long-term performance in BMIs.

Main Methods:

• Proposed an evolutionary constructive and pruning neural network with error feedback (ECPNN-EF).
• Utilized a partially connected topology for decoding neural firing rates into rat forelimb movement.
• Incorporated error feedback as an additional input to compensate for functional mapping changes.

Main Results:

• The ECPNN-EF demonstrated significantly higher daily decoding performance.
• The ECPNN-EF exhibited smaller daily variability in decoding performance.
• Error feedback and partially connected topology were key to improved stability.

Conclusions:

• The ECPNN-EF effectively handles within- and across-day changes in neural recording conditions.
• Error feedback compensates for performance decreases caused by changing conditions.
• This approach enhances long-term decoding stability in BMIs with limited training data.