Related Experiment Video
Updated: Jun 23, 2026
10:51
An Experimental Platform to Study the Closed-loop Performance of Brain-machine Interfaces
Published on: March 10, 2011
Exploring Synergies in Brain-Machine Interfaces: Compression vs. Performance
Luis H Cubillos1,2,3, Madison M Kelberman3, Matthew J Mender3
1Neuromuscular and Rehabilitation Robotics Laboratory (NeuRRo Lab), Physical Medicine and Rehabilitation, Michigan Medicine, Ann Arbor, MI, USA.
Restorative Neurology and Neuroscience
|June 22, 2026
Summary
Muscle synergies and dimensionality reduction techniques like PCA, NMF, and dPCA did not improve brain-machine interface performance for decoding complex movements in primates. Data compression was achieved, but denoising and generalization benefits were not observed.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Robotics
Background:
- Assistive technologies for neurological injuries face challenges decoding multi-degree-of-freedom (DoF) movements.
- Intracortical brain-machine interfaces (iBMIs) offer natural control but struggle with higher DoF movements.
- Muscle synergies are theorized to simplify complex movements by linking muscle activations.
Purpose of the Study:
- To evaluate if muscle synergies enhance iBMI performance in non-human primates.
- To assess if dimensionality reduction techniques (PCA, dPCA, NMF) improve decoding and generalization for implanted recordings.
- To determine if synergies offer a cleaner control space for linear decoding.
Main Methods:
- Applied principal component analysis (PCA), demixed PCA (dPCA), and non-negative matrix factorization (NMF) to neural and muscle data.
- Tested iBMI performance in non-human primates performing a two-DoF finger task.
- Evaluated data compression, denoising, and cross-task generalization capabilities of synergy extraction methods.
Main Results:
- All dimensionality reduction methods effectively compressed neural and muscle data with minimal loss in decoding accuracy.
- No significant improvement in decoding performance was observed through denoising.
- Synergy extraction did not enhance decoder generalization across different tasks.
- Linear decoding performance was not improved by leveraging muscle synergies.
Conclusions:
- Dimensionality reduction aids data compression for iBMIs but does not inherently improve decoding via synergy extraction in this study.
- Extracting muscle synergies alone did not create an advantageous control space for linear decoding.
- Further research with larger datasets and more muscle recording channels is needed to explore synergies as an optimal control framework for iBMIs.
