SMALL SCALE HETEROGENEITY OF HF-LFP AIDS HIGH ACCURACY MOVEMENT DECODING

Many neural signals have been used to decode movement. Local field potentials (LFP), for example, exhibit superior signal stability over time relative to single unit activity (SUA). This is thought, however, to come at the expense of high correlation on small spatial scales. In the present work, we show how high frequency LFP (HF-LFP) serves as a good middle-ground between SUA and LFP for movement decoding. Since recent work has shown HF-LFP is correlated with population spiking dynamics, we hypothesized that much of the heterogeneity of SUA might persist at the level of HF-LFP. To this end, we analyzed neural recordings from two adult male rhesus macaque monkeys (Macaca mulatta) while performing a reach-to-grasp task. Neural data was recorded via Utah arrays, implanted in the left/right ventral premotor cortices (PMv), the left/right dorsal premotor cortices (PMd), and the left primary motor cortex (M1). Spectral electrode amplitude and pairwise Pearson correlation were calculated from band-passed versions of the signal, effectively defining eight frequency bands, ranging from 4-500 Hz. We then used machine learning classifiers to determine decoding accuracy of various states of the reach-to-grasp task ranging from motor preparation to execution and reward retrieval. Our results indicate that intra-region HF-LFP heterogeneity varies by region and motor state. Our results also demonstrate the ability of HF-LFP to capture complex, intra-region, population level dynamics. High heterogeneity was associated with higher motor decoding performance and, surprisingly, heterogeneity alone was sufficient to decode motor states with high accuracy.