Spiking Neural Network Decoders of Finger Forces from High-Density Intramuscular Microelectrode Arrays

Assistive technologies for restoring naturalistic finger control require continuous and robust decoding of motor intent, with high accuracy and low latency. Here, we present a spike-based decoding framework that exploits the dynamics of spiking neural networks (SNNs) to efficiently process motor unit activity extracted from high-density intramuscular microelectrode arrays. Using this framework, we demonstrate simultaneous and proportional decoding of individual finger forces from motor unit spike trains during isometric contractions at 15% of maximum voluntary contraction. We systematically evaluated the properties of different SNN decoder configurations, comparing two possible input modalities: physiologically grounded motor unit spike trains and spike-encoded intramuscular EMG signals. Through this comparison, we determined the trade-offs between decoding accuracy, memory footprint, and robustness to input errors. Our results show that lean shallow SNNs are sufficient to decode finger-level motor intent with competitive accuracy, while operating, with minimal memory requirements and without the need for external pre-processing modules. This work provides a practical blueprint for integrating compact, low-power and low-latency SNNs into finger-level force decoding systems, demonstrating how the choice of input representation can be strategically tailored to meet application-specific requirements for accuracy, robustness, and memory efficiency.