Neuron-Aware Brain-to-Computer Communication for Wireless Intracortical BCI

Intracortical brain-computer interfaces (iBCIs) promise revolutionary clinical and research applications. State-of-the-art iBCIs rely on high-density (HD) microelectrode arrays (MEAs) to sense massive neuronal populations. However, HD MEAs are bandwidth-demanding, posing a significant challenge for wireless iBCIs. Prior iBCI systems have relied on compression to reduce neural signal bitrate. Unfortunately, existing schemes are blind to neurons’ signal characteristics, resulting in poor compression efficiency and severe degradation in iBCI performance.

This paper explores a neuron-aware approach to the design of efficient brain-to-computer communication systems. We present NeuroZip, a neural signal compression scheme that significantly reduces bitrate without compromising neural features, enabling various wireless iBCI applications to track neurons under limited bandwidth. To achieve this, NeuroZip first models and analyzes the complex feature space of HD neural signals, and then embraces neuron-awareness into an efficient genetic search algorithm that can quickly converge to an optimal compression strategy despite the large solution space yielded by HD MEA’s high microelectrode count. Preliminary experiments conducted on real neural datasets show that, compared to neuron-blind schemes, NeuroZip reduces bandwidth by up to 2.2x under the same error constraint, or reduces error rate by up to 8x under the same bandwidth. Further experiments demonstrate NeuroZip imposes minimal impacts on downstream iBCI tasks, limiting the increase of error rate within 2.4% for three representative iBCI applications.

© 2024 Copyright held by the owner/author(s).