Estimation of dynamic functional connectivity using multiplication of temporal derivatives

Abstract

Functional connectivity provides an informative and powerful framework for exploring brain organization. Despite this, few statistical methods are available for the accurate estimation of dynamic changes in functional network architecture. To date, the majority of existing statistical techniques have assumed that connectivity structure is stationary, which is in direct contrast to emerging data that suggests that the strength of connectivity between regions is variable over time. Therefore, the development of statistical methods that enable exploration of dynamic changes in functional connectivity is currently of great importance to the neuroscience community. In this paper, we introduce the 'Multiplication of Temporal Derivatives' (MTD) and then demonstrate the utility of this metric to: (i) detect dynamic changes in connectivity using data from a novel state-switching simulation; (ii) accurately estimate graph structure in a previously-described 'ground-truth' simulated dataset; and (iii) identify task-driven alterations in functional connectivity. We show that the MTD is more sensitive than existing sliding-window methods in detecting dynamic alterations in connectivity structure across a range of correlation strengths and window lengths in simulated data. In addition to the temporal precision offered by MTD, we demonstrate that the metric is also able to accurately estimate stationary network structure in both simulated and real task-based data, suggesting that the method may be used to identify dynamic changes in network structure as they evolve through time.