Enhancing information extraction from field potentials in electrophysiology studies

Abstract

Multi-channel recordings from the brain serve as the primary approach to address various mechanistic and behavioral questions about neural activity and the underlying circuitry. While electrodes can detect neural activity at a distance from their source generators, activity from other concurrently active neural sources is also volume conducted to the electrodes, thus creating linear dependencies among the channels. These dependencies pose challenges in discerning specific communication patterns between different neuronal populations, and their behavioral implications. In this study, we demonstrate the capability of a marked point process (MPP) representation of channel activity, focusing on oscillatory bursts, to curtail the effects of volume conduction noise. By characterizing the localized spectral information within oscillatory bursts, we achieved a[~] 45% reduction in channel correlations across three recording modalities of field potentials (electroencephalography, electrocorticography, and local field potentials). We further provide evidence that the implemented sparse representation preserves both behavioral and causal information in the signal. We illustrate our findings with two examples: 1) retention of finger-level movement information in field potentials recorded from humans, demonstrated using a simple online classifier, and 2) retention of top-down connectivity information between the prefrontal and motor cortices of behaving rats. Overall, our results underscore the novelty of using a marked point process representation of oscillatory bursts to concisely encode behavioral and connectivity information while attenuating the effects of volume conduction from the causal signal sources. Author summaryMulti-channel recordings from the brain are critical to understanding how different brain regions communicate and drive behavior. However, the electrodes that detect neural activity also pick up signals from other active neural sources, leading to misleading linear dependencies among the channels. In our study, we present a novel approach using a marked point process (MPP) representation of channel activity using oscillatory bursts, to minimize the interference caused by volume conduction noise. By isolating the spectral characteristics within these oscillatory bursts, we achieved a[~] 45% reduction in channel correlations across three different types of field potential recordings: electroencephalography (EEG), electrocorticography (ECoG), and local field potentials (LFPs). Importantly, our method not only reduces noise but also retains critical behavioral and causal information in the neural signals. Our findings reveal that a marked point process representation of oscillatory bursts provides a more precise and noise-resilient representation of neural activity. This advancement has significant implications for both understanding brain function and improving the accuracy of brain-machine interfaces.