Spatio-temporal patterns in multi-electrode array local field potential recordings

This paper presents a method for the detection of traveling waves of activity in neural recordings from multielectrode arrays. The method converts local field potential measurements into the phase domain and fits a series of linear models to find planar traveling waves of activity. Here I present the new approach in the context of the previous work it extends, apply the approach to data from neural recordings from a single animal, and verify the success of the method on simulated data. 1 Background and Overview Current technology allows for neural recording on many scales, each with trade offs in terms of spatial resolution, temporal resolution, and recording target. A single extracellular electrode has high temporal resolution and relatively high spatial precision. Electrode arrays, by recording from spatially structured collections of electrodes simultaneously, supplement this with some information on spatio-temporal patterns. Electrode arrays may therefore be a useful recording technique for exploring local network dynamics in vivo. The signal recorded by a single electrode contains high frequency electrical signals from nearby spiking neurons and a low frequency signal referred to as the local field potential (LFP). The LFP is believed to be a measure of the summed synaptic activity in a radius around an electrode (Katzner et al., 2009). Previous work in the Hatsopoulos lab at the University of Chicago (Rubino et al., 2006) suggests that LFP recorded over an array of electrodes can reveal spatio-temporal patterns in activity. Specifically, it can reveal traveling waves of oscillation. Traveling waves are widely discussed in the neuroscience literature (Delaney et al., 1994; Huang et al., 2004; Manjarrez et al., 2007; Wu et al., 2008; Xu et al., 2007; Reimer et al., 2010). Propagating waves of oscillation have been reported in the visual, auditory, olfactory and somatosensory systems as well as in motor cortex, hippocampus, and spontaneous cortical activity. Theoretical work reported in Ermentrout and Kleinfeld (2001) demonstrates that the traveling wave patterns reported in experimental studies are likely to arise in realistic neural networks. Even so, the possible computational purpose, if any, of traveling waves remains unclear. Determining their significance, or even relegating them to the realm of the epiphenomenal, will require advances in the methodology used to detect and describe them. Currently, there is no generally established methodology for detecting and describing traveling waves. Many studies rely on visual inspection of Voltage Sensitive Dye (VSD) images. Several recent papers using LFP electrode recordings make a start on developing formal wave detection techniques. Lubenov and Siapas (2009) used regression methods on a collection of electrodes, and Rubino et al. (2006) used phase derivative calculations on electrode array 1 ar X iv :1 50 1. 00 23 0v 1 [ qbi o. N C ] 1 J an 2 01 5 data. This project synthesizes ideas from these two papers (applied to the data from the later) to develop methodology that allows for statistically rigorous single trial analysis of LFP array data. This research analyzes data from the Hatsopoulos Lab at the University of Chicago. I first replicated the results reported in Rubino et al. (2006) regarding the detection of traveling waves. This previous methodology was able to detect the presence of wavelike activity and capture its broad characteristics, but was unable to provide a single trial description of activity. I made several improvements to the methodology, allowing for single trial analysis. I also demonstrated that wavelike activity is persistent rather than occurring in short isolated time periods as previously reported. Finally, I created artificial data to test both the original and improved methodology under varying noise conditions, demonstrating the efficacy of my improvements. The next two sections of this paper are introductory. Section 2 states the model for a planar wave of oscillation. Section 3 describes the data being modeled. The following section discusses the results of previous analysis, while section 5 describes my analysis and results in detail. Finally, section 6 describes the conclusions from artificial data analysis.