Sparse Bayesian inference methods for decoding 3D reach and grasp kinematics and joint angles with primary motor...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Motor Control

Background:

• Decoding movement intentions from neural activity is crucial for brain-computer interfaces.
• Understanding how the primary motor cortex (M1) represents movement parameters is key to motor control research.

Purpose of the Study:

• To apply sparse Bayesian inference methods for decoding three-dimensional (3D) reach-to-grasp movements.
• To identify the most effective decoding model and assess the representation of joint angles in M1.

Main Methods:

• Recordings of primary motor cortical (M1) ensembles from rhesus macaque.
• Application of variational Bayes (VB) inference with automatic relevance determination (ARD) for variable selection.
• Comparison of three linear and nonlinear decoding models, including sparse Bayesian linear regression.

Main Results:

• The sparse Bayesian linear regression model demonstrated the best decoding performance across various objects and target locations.
• Sensitivity analysis of M1 units in decoding was performed.
• Evaluation of proximal and distal joint angle representations in population decoding.

Conclusions:

• Sparse Bayesian inference, particularly with ARD, is effective for decoding complex movements from M1 ensembles.
• M1 ensembles primarily encode proximal joint angle information rather than distal information during reach-to-grasp tasks.