Detecting neural-state transitions using hidden Markov models for motor cortical prostheses.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Neural prosthetic interfaces decode arm reaching activity for device control.
• Current systems assume known temporal boundaries between neural activity epochs.
• Accurate decoding relies on differentiating baseline, planning, and perimovement neural signals.

Purpose of the Study:

• Develop a technique to automatically differentiate neural activity epochs (baseline, plan, perimovement).
• Enable autonomous operation of neural prosthetic interfaces by identifying neural state transitions.
• Improve brain-directed control of external devices by accurately segmenting neural data.

Main Methods:

• Utilized a hidden Markov model (HMM) as a generative model for neural activity.
• Employed a state-dependent Poisson firing model within the HMM framework.
• Detected transitions between neural activity epochs using a posteriori HMM state probabilities.

Main Results:

• Successfully detected transitions from baseline to plan epochs, even without behavioral changes.
• Achieved decoding accuracy comparable to maximum-likelihood estimators for movement targets.
• Demonstrated the HMM's ability to detect neural transitions for previously unseen movement targets.

Conclusions:

• The HMM technique automatically identifies distinct neural activity epochs crucial for prosthetic control.
• This method allows neural prosthetic interfaces to operate autonomously, adapting to changing neural states.
• Enables more robust and intuitive brain-computer interfaces for assistive technologies.