Exploring Synergies in Brain-Machine Interfaces: Compression vs. Performance.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Severe neurological injuries limit mobility, necessitating advanced assistive technologies.
• Current assistive technologies, including intracortical brain-machine interfaces (iBMIs), struggle with decoding complex, multi-degree-of-freedom (DoF) movements.
• Muscle synergies, a theoretical neural simplification strategy, have shown promise in non-invasive applications for noise reduction.

Purpose of the Study:

• To evaluate the effectiveness of muscle synergies and dimensionality reduction techniques in enhancing iBMI performance for high-DoF movements.
• To determine if principal component analysis (PCA), demixed PCA (dPCA), and non-negative matrix factorization (NMF) can improve decoding accuracy and generalization in implanted iBMIs.
• To assess the utility of these methods for compressing and denoising neural and muscle data in non-human primates.

Main Methods:

• Applied PCA, dPCA, and NMF to neural and muscle recordings from non-human primates performing a two-DoF finger task.
• Evaluated the impact of these dimensionality reduction techniques on decoding accuracy and generalization across tasks.
• Assessed the ability of these methods to compress and denoise data for iBMI applications.

Main Results:

• All tested dimensionality reduction methods (PCA, dPCA, NMF) effectively compressed neural and muscle data with minimal loss in decoding accuracy.
• None of the dimensionality reduction techniques improved decoding performance through denoising.
• No significant enhancement in decoder generalization across different tasks was observed using these methods.

Conclusions:

• While dimensionality reduction aids in data compression for iBMIs, it does not inherently improve decoder performance or generalization on its own.
• Muscle synergies, when analyzed via PCA, dPCA, or NMF, may not represent the optimal control framework for enhancing iBMI robustness.
• Further research is needed to explore alternative approaches for improving decoder performance and generalizability in iBMI applications for individuals with neurological injuries.