Challenges and opportunities for next-generation intracortically based neural prostheses

Abstract

Neural prosthetic systems aim to help disabled patients by translating neural signals from the brain into control signals for guiding computer cursors, prosthetic arms, and other assistive devices. Intracortical electrode arrays measure action potentials and local field potentials from individual neurons, or small populations of neurons, in the motor cortices and can provide considerable information for controlling prostheses. Despite several compelling proof-of-concept laboratory animal experiments and an initial human clinical trial, at least three key challenges remain which, if left unaddressed, may hamper the translation of these systems into widespread clinical use. We review these challenges: achieving able-bodied levels of performance across tasks and across environments, achieving robustness across multiple decades, and restoring able-bodied quality proprioception and somatosensation. We also describe some emerging opportunities for meeting these challenges. If these challenges can be largely or fully met, intracortically based neural prostheses may achieve true clinical viability and help increasing numbers of disabled patients.

Fig. 1

Three key challenges for intracortically based neural prostheses. (a) Notional sketch of the performance versus burden design space with the goal of complete functional restoration shown in green. (b) Notional sketch of how single-neuron-based information from intracortical electrode arrays varies (solid red line) and declines over time (dashed red lines, one for each hypothetical array implant) with the goal of high, stable, and long lasting information availability shown in green. (c) Block diagram illustrating the current capability to read out electrical neural signals, “write in” information using eICMS, and highlighting the need for high-fidelity information write in potentially enabled by oICMS (green) of optogenetically transfected neurons. A bidirectional prosthetic example is shown wherein neural activity is decoded to guide a prosthetic arm/hand (blue), and signals from electronic pressure sensors (magenta) are encoded into optical pulses (green) to provide the artificial sense of touch of an object (brown).

Fig. 2

Action potential waveform instability and prosthetic decode performance stability. (a) Action potential waveforms from one example electrode at various times during a two-week wireless recording session. The electrode is one of 96 on the electrode array (Blackrock Microsystems) implanted in PMd of rhesus Monkey L. (b) Mean peak-to-peak voltages from neurons on the 96 electrodes that had V_pp > 200 µV on the first day of recording (i.e., selection criterion). These neurons came from 35 (of the 96) electrodes and reflect the “best neurons” on the array if relying on individual neurons and spike sorting for operating a prosthesis. The array was implanted in M1 in Monkey J. (c) Percent success classifying one of four reach directions with a maximum likelihood decoder using 500 ms of post-go cue threshold-crossing events (−4.5 × V_{R M S}). This analysis was performed using threshold crossings from all 96 electrodes in M1 of Monkey J, but excluding approximately ten electrodes as the threshold-crossing rates on these electrodes was below our criterion of 0.2 threshold crossings per second. Threshold-crossing events were confirmed to be of neural origin (single neuron action potentials, or multiunit action potentials often referred to as “hash”) by applying a shape heuristic and by visual inspection. This illustrates the relative stability of information derived from threshold crossings (panel c and Fig. 1(b), green line), and contrasts with the relatively unstable information available when reliant on individual neurons and spike sorting (panels a and b and Fig. 1(b), red lines). Adapted from [59].

Fig. 3

Response of an example ChR2 transfected neuron and an eNpHR2.0 transfected neuron in cerebral cortex of a rhesus monkey. (a) Raster plot and peri-stimulus time histogram (PSTH) showing an increase in action potential emission rate during blue light illumination. The neuron is from a site injected with AAV5-hThy-1-ChR2-EYFP. Spontaneous (black) and light-triggered action potential waveforms (blue) are indistinguishable in shape (insets). Monkey D. (b) Raster plot and PSTH showing a decrease in action potential emission rate during green light illumination. The neuron is from a site injected with AAV5-hThy-1-eNpHR2.0-EYFP. Monkey D. Adapted from [87].