Reconstruction of hand, elbow and shoulder actual and imagined trajectories in 3D space using EEG slow cortical potentials

• Published in: Journal of neural engineering Vol. 17; no. 1; pp. 16065 - 16079
• Main Authors: Sosnik, Ronen, Ben Zur, Omer
• Format: Journal Article
• Language: English
• Published: England IOP Publishing 18.02.2020
• Subjects: Adult; Biomechanical Phenomena - physiology; brain-computer interface (BCI); Brain-Computer Interfaces; Elbow - physiology; electroencephalography (EEG); Electroencephalography - methods; Hand - physiology; Humans; Imagination - physiology; imagined 3D arm movements; Male; motion trajectory prediction (MTP); Motor Cortex - physiology; Movement - physiology; multiple linear regression (mLR); Shoulder - physiology; slow cortical potentials (SCPs); Young Adult
• ISSN: 1741-2560; 1741-2552; 1741-2552
• DOI: 10.1088/1741-2552/ab59a7

Objective. The ability to decode kinematics of imagined movement from neural activity is essential for the development of prosthetic devices that can aid motor-disabled persons. To date, non-invasive recording methods, including electroencephalogram (EEG) were used to decode actual and imagined hand trajectory to control neuromotor prostheses, commonly by applying multi-dimensional linear regression (mLR) models to adjust the two temporal signals-neural signal and limb kinematics. It is still debated, however, whether the EEG signal, in general, and slow cortical potentials (SCPs), in specific, hold motor neural correlates. Moreover, it has not yet been tested whether the trajectory of proximal arm joints, i.e. shoulder, can also be reconstructed and if decoding performance is dependent on movement speed and/or position variance. Approach. We predicted hand, elbow and shoulder trajectories in 3D space in time series of both movement types (actual and imagined) of seven subjects using an mLR model, commonly applied for motion trajectory prediction (MTP) and used source localization to detect and compare between brain areas activated during actual and imagined movements for each arm joint. Main results. For all arm joints and movement types, SCPs contributed the most to trajectory reconstruction, and decoding accuracy peaked using neural signals preceding kinematics by 120-210 ms. The average (across subjects) Pearson's correlation coefficient between predicted and actual trajectories ranged 0.24-0.49, 0.41-0.48 and 0.18-0.40 for the hand, elbow and shoulder, respectively, and was significantly higher than chance level (p   <  0.01) for all subjects. For the imagined movements, reconstruction accuracy ranged between 0.09-0.23, 0.20-0.27 and 0.11-0.18 for the hand, elbow and shoulder, respectively, and was significantly higher than chance level (p   <  0.05) for all or some of the arm joints. The model performance was positively correlated with movement speed and negatively correlated with position variance. Source localization suggested that the neural circuits engaged in motor imagery are more diffuse and bilateral; motor imagery was, when compared to movement execution, more associated with recruiting premotor regions and a large area of the left parietal cortex. Significance. Our results demonstrate the feasibility of predicting 3D imagined trajectories of all arm joints from scalp EEG and imply the existence of movement related neural correlates in slow cortical potentials.