Unsupervised Neural Manifold Alignment for Stable Decoding of Movement from Cortical Signals.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Stable decoding of movement parameters from neural activity is essential for brain-machine interfaces (BMIs).
• Neural activity instability over time degrades movement decoding accuracy in BMIs.
• Current manifold stabilization techniques require known subject intentions, limiting practical application.

Purpose of the Study:

• To develop an automatic, unsupervised algorithm for stabilizing brain-machine interfaces (BMIs).
• To determine movement target intention prior to manifold alignment, addressing limitations of existing methods.
• To enhance the accuracy and stability of long-term movement decoding in BMIs.

Main Methods:

• An unsupervised algorithm was developed to determine movement target intention without prior knowledge.
• This algorithm was integrated with dimensionality reduction and manifold alignment techniques.
• The combined method was tested on decoding 2D hand velocity in rhesus macaques during a center-out task.

Main Results:

• The proposed method successfully stabilized decoding across sessions, even with manifold rotation and scaling.
• Decoding performance was evaluated using correlation coefficient and R-squared measures.
• The approach demonstrated superior decoding performance compared to a state-of-the-art unsupervised BMI stabilizer.

Conclusions:

• The developed unsupervised algorithm offers an automatic solution for stable and accurate movement decoding in BMIs.
• This method overcomes the need for known subject intentions, advancing long-term BMI applications.
• The findings contribute to more reliable and robust brain-machine interface systems.