Classifier Performance in Primary Somatosensory Cortex Towards Implementation of a Reinforcement Learning Based Brain...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Advancements in brain-machine interface (BMI) control theory are improving prosthetic limb control.
• Reinforcement learning (RL) is a promising technique for future BMI applications, requiring a reward signal to guide control.
• This reward signal has been observed in cortical structures used for BMI control.

Purpose of the Study:

• To evaluate the ability of common classifiers to detect impending reward delivery in primary somatosensory (S1) cortex.
• To assess classifier accuracy across various conditions to find optimal parameters for classification.
• To determine if S1 cortex can be used to adapt RL-based BMIs to user intent.

Main Methods:

• A nonhuman primate performed a grip force match-to-sample task.
• Several common classifiers were used to detect reward signals in S1 cortex.
• Classifier accuracy was evaluated under diverse conditions and data input parameters.

Main Results:

• S1 cortex demonstrated highly accurate classification of the reinforcement signal across multiple classifiers and parameters.
• Classification accuracy was evident when the animal anticipated reward or reward withholding.
• The accuracy was consistent across various tested conditions.

Conclusions:

• Primary somatosensory cortex accurately classifies reward-related neural signals, crucial for RL-based BMIs.
• These findings support the use of S1 cortex signals for adapting RL-based BMIs to user intent.
• Real-time implementation could enable dynamic prosthetic control that matches user intentions.