DESIGN PRINCIPLES OF A NEUROMOTOR PROSTHETIC DEVICE

Neuromotor prostheses are a type of brain-machine interface (BMI) that seek to extract signals from the central or peripheral nervous system and deliver them to control devices. A brain-machine interface is necessary to detect activity that can be voluntarily modulated for use as a motor control signal. It is generally accepted that electrical potentials are the most valuable sources of information. Neural commands for voluntary movement are essentially issued as electrical signals produced by the spiking (action potentials) and synaptic input of individual neurons; both can be recorded with varying degrees of fidelity and difficulty. The goal is to be able to detect signals that have the largest amount of information about movement and that change about as rapidly as movement commands themselves change. Clearly, recording at the source of the motor commands most readily fulfills these requirements, but indirect recordings of surrogate signals can provide an alternative or supplemental source, if one can learn to make indirect signals mimic motor commands. The decoding methods for use in neuromotor prostheses are the culmination of many years of basic research on the motor system. Whereas recovering movement dynamics and kinematics from neural activity alone comprises a feat of basic science, their use as a control signal marks a shift to applied neuroprosthetics. In this chapter we review mathematical algorithms that have been tested in prototypes of intracortical neuromotor prostheses. ‘Closed-loop’ refers to the situation wherein the subject is provided access to recovered movement information, and is required to use this prediction signal in a behaviorally useful manner. This access may be afforded visually (neurally derived cursor trajectories), mechanically (as in stimulation of muscles via implanted electrodes), or any number of output devices. We will consider several features that are unique to the closed-loop context of online control, including those specific to use in paralyzed human patients. We consider here the advantages and disadvantages of field potentials and spikes; in the final section of the chapter, we argue that a principled combination of all available information channels, processed by a multiplicity of decoding algorithms, will result in the most effective neuromotor prosthesis.