Design principles for noninvasive brain-machine interfaces

With the advent of sophisticated prosthetic limbs, the challenge is now to develop and demonstrate optimal closed-loop control of the these limbs using neural measurements from single/multiple unit activity (SUA/MUA), electrocorticography (ECoG), local field potentials (LFP), scalp electroencephalography (EEG) or even electromyography (EMG) after targeted muscle reinnervation (TMR) in subjects with upper limb disarticulation. In this paper we propose design principles for developing a noninvasive EEG-based brain-machine interface (BMI) for dexterous control of a high degree-of-freedom, biologically realistic limb.

I. Introduction

Electromyography (EMG)-based systems have shown reasonably reliable 7-degrees-of-freedom (DOF) control of a prosthetic limb using EMG after TMR - a surgical technique pioneered by Dr. Kuiken involving the transfer of residual nerves in the amputated arm to the remaining muscle, which then provide EMG signals that correlate to the original nerve functions allowing a virtual or physical prosthetic arm to respond directly and more naturally to the brain signals [1]–[2]. Some critical challenges of this approach concern the stability of EMG recordings, interference from muscles controlling remaining joints, effects of tissue loading, control of fine dexterous movements, and the cognitive burden of operating the device [1]. Thus, it is desirable to develop noninvasive neural interfaces that directly use brain signals, such as scalp electroencephalography (EEG), to control fine dexterous movements.