Design and validation of a real-time spiking-neural-network decoder for brain–machine interfaces

Abstract

Objective Cortically-controlled motor prostheses aim to restore functions lost to neurological disease and injury. Several proof of concept demonstrations have shown encouraging results, but barriers to clinical translation still remain. In particular, intracortical prostheses must satisfy stringent power dissipation constraints so as not to damage cortex. Approach One possible solution is to use ultra-low power neuromorphic chips to decode neural signals for these intracortical implants. The first step is to explore in simulation the feasibility of translating decoding algorithms for brain-machine interface (BMI) applications into spiking neural networks (SNNs). Main results Here we demonstrate the validity of the approach by implementing an existing Kalman-filter-based decoder in a simulated SNN using the Neural Engineering Framework (NEF), a general method for mapping control algorithms onto SNNs. To measure this system’s robustness and generalization, we tested it online in closed-loop BMI experiments with two rhesus monkeys. Across both monkeys, a Kalman filter implemented using a 2000-neuron SNN has comparable performance to that of a Kalman filter implemented using standard floating point techniques. Significance These results demonstrate the tractability of SNN implementations of statistical signal processing algorithms on different monkeys and for several tasks, suggesting that a SNN decoder, implemented on a neuromorphic chip, may be a feasible computational platform for low-power fully-implanted prostheses. The validation of this closed-loop decoder system and the demonstration of its robustness and generalization hold promise for SNN implementations on an ultra-low power neuromorphic chip using the NEF.

Predicting voluntary movements from motor cortical activity with neuromorphic hardware

“…The conventional Kalman filter has repeatedly been shown to be successful in decoding smooth hand movement trajectories from motor cortical activity. The authors simulated their network on a computer fast enough to achieve decoding in real time and they successfully tested this approach in an experiment where the hand movement trajectories were accurately decoded in real time [22] (a closed-loop scenario). As input to their network the authors used the multi-unit spike count (or firing rate) estimated in 50 ms time bins from each channel recorded with a so-called Utah array (a 96-channel electrode system).…”

“…The temporal resolution of the simulation is 15-25 ms. As input to the rate neurons the authors provided the spike count (or firing rate) estimated in bins that use resolution of the simulation (15-25 ms) from multi-unit activity of 96 channels simultaneously recorded with a Utah array in a closed-loop scenario. These experiments were performed in the same lab and with the same monkeys as in [22].…”

“…The most appropriate classifier to be used under such a paradigm is a spiking neural network, which draws inspiration from biological brains and uses spikes to relay information between its neurons (units). The use of a spiking neural network decoder has previously been suggested by Dethier and colleagues [21], [22], who have successfully simulated a spiking neural network that mimicked a Kalman filter to decode firing rates of multiunit activity in a closed-loop scenario.…”

“…Therefore, they are in principle well suited for embedded and portable applications. Despite these advantages, the use of spiking neuromorphic systems to decode single neuron activity in real-time has only rarely been considered in the literature (i.e., in [22]), and, to our knowledge, no actual implementation has yet been published.…”

A High-Sensitivity Fully Passive Neurosensing System for Wireless Brain Signal Monitoring

“…Another issue is the heat generated by the employed dense electronics [8] [9]. This heating might disturb normal brain operation and damage the cerebral tissue.…”