Characterization of phase estimation techniques to guide phasic stimulation for brain machine interfacing

Abstract

While much work on brain-machine interfaces (BMI) focuses on decoding neural signals to drive external plants (e.g. a robotic arm), there is increasing interest in the delivery of central neural stimulation, for applications including feedback control (e.g. robotic arm proprioception), or the replacement of impaired sensory function (e.g. haptic prostheses). In many cases, the effects of stimulation can be influenced by pre-stimulus neural activity. Specifically, oscillatory activity contributes significantly to modulating neural responsiveness, and effective use of cortical stimulating BMIs may require phase dependent input. This work compares Hilbert transform and extended Kalman filter (EKF) methods for controlling stimulation at targeted phases of ongoing neural oscillations. Accuracy of stimulation timing and computational latency were the primary criteria used to assess both approaches. Algorithmic performance was evaluated on signals ranging from noisy sinusoids to previously recorded cortical local field potentials. Characterizing the abilities and limitations of these two filtering techniques is a step towards the development of user-defined phase stimulation in BMIs.