Over 600,000 Americans have severely impaired motor function from disorders including spinal cord injury, amyotrophic lateral sclerosis, pontine stroke, and cerebral palsy. A brain-machine interface (BMI) could enable locked-in or tetraplegic patients to communicate and interact with their environment. Two crucial decisions in designing a BMI are (1) what type of brain signals to use as inputs to a controller and (2) what methods to use to decode those signals. Most BMIs have used either noninvasive scalp EEG recordings or invasive intracortical recordings of single- or multi-neuron spikes as control inputs. A few have used subdural or intracortical local field potentials (LFPs). However, no group has yet systematically compared these signals in motor cortex for use in BMI applications. This proposal's first goal is to assess the relative performance of spikes and field potentials (both intracortical and epidural) as control inputs for a variety of movement-related outputs. Epidural field potentials (EFPs) are intermediate in invasiveness, signal quality, stability and spatial resolution compared with existing scalp, subdural, and intracortical recordings, and thus represent an unexplored middle ground. This proposal's second goal is to evaluate linear and nonlinear techniques-including several novel to BMI applications-for both decoding data and reducing the inherently large dimensionality of data from multiple neural signals. The primary hypotheses of the proposed project are (1) that spikes will perform better in decoding more complex movement-related outputs, but that field potentials may perform similarly on decoding simpler outputs, and (2) that nonlinear decoders and dimensionality-reduction techniques may provide improved accuracy over linear methods.
The specific aims to address these hypotheses are 1) to evaluate single neuron spikes as inputs to decoders of movement- related outputs, 2) to develop a novel epidural multi-electrode recording technique in the macaque monkey, and 3) to evaluate field potential signals as inputs to decoders of movement-related outputs.
Aims 1 and 3 will involve application of dimensionality-reduction algorithms (e.g., independent components analysis, Isomap) and decoding algorithms (system identification, neural networks, support vector machines) to both spikes and field potentials.
Aim 2 will entail using a computer model and spatial spectral analysis to optimize the epidural electrode array design. This project will provide the first comparison of spikes, LFPs and EFPs as inputs for identical BMI output applications. The supervision of Drs. Lee Miller and W. Zev Rymer, with additional guidance from Drs. Simon Levine, Jonathan Wolpaw and Nicholas Hatsopoulos, will provide the principal investigator with expertise in recording and processing both spikes and field potentials for BMI applications using a variety of state-of-the- art techniques. A comprehensive career development plan including clinical and research mentoring, seminars, and courses, will foster the candidate's transition into an independent physician-scientist.
Showing the most recent 10 out of 13 publications
