Single trial decoding of visual attention from local field potentials in the primate lateral prefrontal cortex

摘要

It has recently been shown that attention can be decoded from the spiking activity of ensembles of neurons in area 8A of the monkey lateral prefrontal cortex (Tremblay et al., 2015). As spikes are believed to reflect the output of a given brain region and local field potentials (LFPs) its input and local activity, we used a decoding approach to investigate the role of area 8A in visual attention. To do so, we used simultaneously recorded LFPs from chronically implanted multi-electrode arrays to decipher the target of visual attention. However, when attempting to decode the information content of an LFP signal, conventional filtering techniques (low-pass at 250 Hz) are insufficient in segregating spikes components from other LPF sources. Indeed, multiple studies have shown that spike waveforms recorded from the same electrode contaminate LFP signals above 80 Hz (Zanos et al., 2012), biasing the decoded information. Using a previously published spike removal algorithm (Zanos et al., 2011), we compared the decoding performances of spike-free LFPs and spiking data. Our results demonstrate that LFPs in the mid to high gamma range ( 80 Hz) offer comparable performance to spikes in decoding accuracy, while the lower frequencies are unreliable. We further show that the information content of high-frequency LFPs is entirely redundant with the one contained in locally recorded action potentials. These results first demonstrate that LFP signals can be used to decode the allocation of attention across the visual field. Secondly, recent work by Ray and Maunsell (2011) proposed that the higher LFP frequencies ( 80 Hz) reflect the neuronal firing around the electrode and are independent of lower gamma oscillations. This suggests that our decoding accuracy is only dependent on the local spiking activity of area 8A and not from its input, placing it as a potential source for attentional signals.