Neural Decoding of Synergy-Based Hand Movements Using Electroencephalography

Abstract

The human central nervous system (CNS) effortlessly performs complex hand movements with the control and coordination of multiple degrees of freedom. It is hypothesized that the CNS might use kinematic synergies to reduce the complexity of movements, but how these kinematic synergies are encoded in the CNS remains unclear. In order to investigate the neural representations of kinematic synergies, scalp electroencephalographic (EEG) signals and hand kinematics were recorded from ten subjects during six representative types of hand grasping. Kinematic synergies were obtained from recorded hand kinematics using singular value decomposition. The recorded kinematics were then reconstructed using weighted linear combinations of synergies, and the optimal weights were computed using optimal linear estimation. Using EEG spectral powers as neural features, a multivariate linear regression model was trained on the weights of the kinematic synergies. Using this model, kinematics from the testing subset of data was decoded from the EEG features with threefold cross-validation. The results show that the weights of kinematic synergies used in a particular movement reconstruction were strongly correlated to EEG features obtained from that movement. EEG features were able to successfully decode synergy-based movements with an average decoding accuracy of 80.1 ± 6.1 % (best up to 93.4 ± 2.3%). These results have promising applications in noninvasive neural control of synergy-based prostheses and exoskeletons.

Keywords

• Electroencephalography
• principal component analysis
• kinematic synergies
• movement primitives
• multivariate linear regression