Optimal Multichannel Artifact Prediction and Removal for Brain Machine Interfaces and Neural Prosthetics
Neural implants that electrically stimulate neural tissue such as deep brain stimulators, cochlear implants (CI), and vagal nerve stimulators are becoming the routine treatment options for various diseases. Optimizing electrical stimulation paradigms requires closed-loop stimulation using simultaneous recordings of evoked neural activity in real time. Stimulus-evoked artifacts at the recording site are generally orders of magnitude larger than the neural signals, which challenge the interpretation of evoked neural activity. We developed a generalized artifact removal algorithm that can be applied in a variety of neural recording modalities. The procedure leverages known electrical stimulation currents to derive optimal filters that are used to predict and remove artifacts. We validated the procedure using paired recordings and electrical stimulation from sciatic nerve axons, high-rate bilateral CI stimulation, and concurrent multichannel stimulation in auditory midbrain and recordings in auditory cortex. We demonstrate a vast enhancement in the quality of recording even for high-throughput multi-site stimulation with typical improvements in the signal-to-noise ratio between 20-40 dB. The algorithm is efficient, can be scaled to arbitrary number of sites, and is applicable in range of recording modalities. It has numerous benefits over existing approaches and thus should be valuable for emerging neural recording and stimulation technologies.