(23) Decoding Kinematic Features from High-Frequency Surface Potentials and Single-Unit Activity in Primate Motor Cortex

Introduction: Decoding motor intent from electrical recordings of brain activity is important for advancing neuroprosthetic technologies for individuals with neurodegenerative disorders or injuries. Technologies to record electrical activity involve either cortical surface electrodes that are placed on the cortical surface and intracortical electrodes that penetrate the cortical tissue. Cortical surface electrodes offer good spatial coverage of the cortical sheet and do not cause damage to the underlying tissue. Intracortical electrodes offer cellular signal resolution in the form of action potential recordings, and resolve signals across the cortical layers. However, directly comparing signal information and how electrode properties impact motor decoding remains poorly understood.

Methods: We recorded surface potentials using a uECoG electrode and recorded action potentials using a Neuropixel electrode. The uECoG electrode had 244 gold contacts (200um size) spaced 750 um in a liquid-crystal polymer substrate. The neuropixel electrode contained 960 TiN contacts (12 um size) spaced 20 um along a 10mm silicon shank. Of the 960 contacts, 384 contacts were selected for recording. We implanted each electrode into the motor cortex of two macaque monkeys (macaca mulatta). We recorded neural activity as each monkey performed a free-reach behavioral task and a center-out behavioral task. We spatially registered the neuropixel recording site to the uECoG recording sites. We measured reach kinematics using three cameras and computed 3D wrist position with computer vision techniques (SLEAP).

Results: We fit a linear regression to predict wrist kinematics from either single-unit activity (SUA) recorded by the neuropixel, Local Motor Potential (LMP) recorded by the uECoG, and high-gamma (70-250 Hz) power recorded by the uECoG. We found that high-gamma power across the uECoG was more correlated with wrist velocity than LMP from uECoG and SUA from the neuropixel. Additionally, the predictive strength of ECoG activity varied by anatomical location, with distinct channels exhibiting higher correlations with kinematics at different phases of the task. This suggests that the broad spatial coverage provided by uECoG captures distributed motor representations, which may be advantageous for decoding applications.

Conclusion: These results indicate surface potentials encode similar information about motor intent to action potentials. Consequently, uECoG may serve as a minimally-invasive electrode interface in neuroprosthetic technologies for individuals with neurodegenerative disorders or injuries.