Nonparametrically estimating dynamic bivariate correlation using visibility graph algorithm

TL;DR

This paper introduces a robust nonparametric method based on the visibility graph algorithm to improve dynamic correlation estimation in neuroscience data, outperforming traditional DCC especially in the presence of outliers.

Contribution

A novel nonparametric DCC method using the weighted visibility graph algorithm, enhancing robustness and applicability in neuroscience time series analysis.

Findings

Better performance on empirical fMRI and LFP datasets

Improved robustness to outliers

Potential to expand analytical tools for brain activity coupling

Abstract

Dynamic conditional correlation (DCC) is a method that estimates the correlation between two time series across time. Although used primarily in finance so far, DCC has been proposed recently as a model-based estimation method for quantifying functional connectivity during fMRI experiments. DCC could also be used to estimate the dynamic correlation between other types of time series such as local field potentials (LFP's) or spike trains recorded from distinct brain areas. DCC has very nice properties compared to other existing methods, but its applications for neuroscience are currently limited because of non-optimal performance in the presence of outliers. To address this issue, we developed a robust nonparametric version of DCC, based on an adaptation of the weighted visibility graph algorithm which converts a time series into a weighted graph. The modified DCC demonstrated better performance in the analysis of empirical data sets: one fMRI data set collected from a human subject performing a Stroop task; and one LFP data set recorded from an awake rat in resting state. Nonparametric DCC has the potential of enlarging the spectrum of analytical tools designed to assess the dynamic coupling and uncoupling of activity among brain areas.