Toward an autonomous brain machine interface: integrating sensorimotor reward modulation and reinforcement learning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Primate Motor Control

Background:

• Cortical sensorimotor control has been studied via kinematics and dynamics.
• Computational models simulate motor cortex learning but require reward signals.
• Reward modulation in the primary sensorimotor cortex is not well-characterized at the neural unit level.

Purpose of the Study:

• To investigate reward expectation modulation in the primary motor cortex (M1) at the level of single neural units.
• To determine if M1 activity can be used to classify rewarding versus nonrewarding trials.
• To explore the potential for using M1 reward information in autonomous brain-machine interfaces.

Main Methods:

• Recorded single units/multiunits and local field potentials in the primary motor cortex (M1) of nonhuman primates (Macaca radiata).
• Analyzed neural activity during reaching movements and passive viewing of reward-predictive cursor motions.
• Developed classification methods to distinguish rewarding from nonrewarding trials based on neural signals.

Main Results:

• Single units and local field potentials in M1 exhibited significant modulation by reward expectation.
• Reward modulation was observed during both active reaching and passive viewing tasks.
• Rewarding versus nonrewarding trials could be classified moment-to-moment using M1 neural activity.

Conclusions:

• Reward expectation influences neural activity in the primary motor cortex (M1).
• M1 contains information that can be decoded to identify trial outcome (rewarding vs. nonrewarding).
• This demonstrates the feasibility of integrating reward information from M1 into autonomous brain-machine interfaces for adaptive decoding.