Optimizing recording depth to decode movement goals from cortical field potentials

Brain-machine interfaces decode movement goals and trajectories from neural activity that is recorded using chronically-implanted microelectrode arrays. Fixed geometry arrays are limited for this purpose because electrodes cannot be moved after implantation, and optimization of the electrode recording configuration requires the re-implantation of a new array. Here, we optimize local field potential (LFP) recordings using a chronically-implanted microelectrode array with electrodes that can be moved after implantation. In a series of recordings, we systematically vary the depth of each electrode in the frontal eye field of a monkey performing eye movements. We find that a decoder predicting movement goals from LFP activity on 32 electrodes provides information rates as high as 5.0 bits/s and that performance varies significantly with recording depth. These results indicate that recording depth is a critical parameter for the performance of LFP-based brain-machine interfaces that decode movement goals.