Neural Decoding Forelimb Trajectory Using Evolutionary Neural Networks with Feedback-Error-Learning Schemes.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Neural decoders in brain-machine interfaces (BMIs) face challenges due to time-varying neural activity and kinematic parameter mappings.
• Traditional decoders often require extensive training data to adapt to these changes, limiting their real-world applicability.
• Developing decoders robust to mapping variations with limited data is crucial for advancing BMI technology.

Purpose of the Study:

• To propose and validate an evolutionary neural network with error feedback (ECPNN-EF) as a robust neural decoder for BMIs.
• To enhance decoder performance by incorporating previous error information to adapt to changing neural-to-kinematic mappings.
• To assess the efficacy of ECPNN-EF using limited training data.

Main Methods:

• An evolutionary neural network with error feedback (ECPNN-EF) was developed as a novel neural decoder.
• The decoder was trained using two days of data from a rat's forelimb movement task.
• Performance was evaluated by testing the decoder on ten days of data, comparing it against a standard recurrent neural network.

Main Results:

• The ECPNN-EF decoder demonstrated significantly higher performance in reconstructing rat forelimb movements compared to a standard recurrent neural network.
• The proposed decoder showed robustness to changes in neural-to-kinematic mappings even when trained with only two days of data.
• Error feedback mechanism proved effective in improving decoder adaptability and performance over time.

Conclusions:

• ECPNN-EF offers a robust solution for neural decoding in BMIs, effectively handling variations in neural-to-kinematic mappings.
• The proposed method significantly reduces the amount of training data required, making BMIs more practical.
• This research paves the way for more adaptive and efficient brain-machine interfaces.