Comparison of long-term decoding stability of ultra high frequency band local field potentials (>500Hz) and spike signals in primate motor cortex

It has been widely reported that the local field potentials (LFPs) can be a useful signal for neural decoding as well as the spike in the brain-machine interfaces (BMIs) studies. However, most researchers mainly focused on the low-frequency bands LFPs and only a few studies used on the ultra high frequency band (>500 Hz) LFPs (uhfLFPs). In this study, we compare the long-term decoding stability of uhfLFPs and multi-unit activities (MUAs) when a male macaque monkey was performing a center-out task. LFPs and MUAs were collected from dorsal premotor cortex of the monkey with a microelectrode array. Neural data were recorded from 4 to 8 months post-implantation, during which period a part of the spikes were lost. The power spectrum of uhfLFPs (>500 Hz) and the firing rate of MUAs were calculated as the features of neural signals. A K-nearest neighbor classifier and a Kalman filter were used to decode the direction and the trajectories of wrist movement respectively. Our results showed that uhfLFPs got a better performance in long-term decoding in both discrete and continuous decoding. It infers that uhfLFPs could be an alternative source of control signals in long-term BMIs study.