Electrocorticography (ECoG)

In collaboration with the Laboratory of Behavioral and Cognitive Neurology and as part of the Stanford Human Intracranial Cognitive Electrophysiology Program (SHICEP), NPTL investigates the use of Electrocorticography (ECoG) electrode grids for neural prosthetic applications.  The ECoG grids are placed on top of the cortex, under the dura, as shown in the image to the right.  We record from two sizes of electrodes, macro (in white) and micro (in red), which provide a tradeoff between recording noise and spatial specificity.  

Macro and Micro ECoG Electrodes on the Cortex

Electrode PlacementHand posture data acquisition


We are interested in developing systems that permit restoration of arm and hand function, and so we record from electrodes in motor and sensory areas of the brain.  We ask our research participants to make specific hand movements that we record with a data glove.  As they make these movements, we record coincident neural activity on specific electrodes.

Gamma response to hand movementsEach panel to the right plots multiple single trial neural responses from an electrode. The responses plotted are power in the high gamma band (66-114 Hz) and are color coded based on movement type; responses to thumb movements (blue) and ring finger movements (green) are shown.  All movements were intitiated from a rest position.  The red line indicates the time of movement onset.  Note the presensence of movement type specific activation across trials.

Since we record from multiple electrodes simulataneously (up to 128), we wish to get a sense for how the neural activity evolves in aggregate.  The figure to the right plots a representation of singles trials (each trace is a single trial color coded to highlight particular behavioral epochs).  The two factors are linear projections of the gamma activation from 64 channels, projection axes are chosen via factor analsysis.  The trajectories on the left were generated from neural activity during trials for which the participant was instructed to make a pinching movement; those on the right were similarly generated from trials for which the participant moved into a pointing hand posture.

With these data, we develop neural decoding algorithms for the real time control of robotic protheses.

Low-D representation of neural activity