Rethinking Measures of Functional Connectivity via Feature Extraction

Abstract

Functional magnetic resonance imaging (fMRI)-based functional connectivity (FC) commonly characterizes the functional connections in the brain. Conventional quantification of FC by Pearson's correlation captures linear, time-domain dependencies among blood-oxygen-level-dependent (BOLD) signals. We examined measures to quantify FC by investigating: (i) Is Pearson's correlation sufficient to characterize FC? (ii) Can alternative measures better quantify FC? (iii) What are the implications of using alternative FC measures? FMRI analysis in healthy adult population suggested that: (i) Pearson's correlation cannot comprehensively capture BOLD inter-dependencies. (ii) Eight alternative FC measures were similarly consistent between task and resting-state fMRI, improved age-based classification and provided better association with behavioral outcomes. (iii) Formulated hypotheses were: first, in lieu of Pearson's correlation, an augmented, composite and multi-metric definition of FC is more appropriate; second, canonical large-scale brain networks may depend on the chosen FC measure. A thorough notion of FC promises better understanding of variations within a given population.open Scientific RepoRtS | (2020) 10:1298 | https://doi.org/10.1038/s41598-020-57915-wwww.nature.com/scientificreports www.nature.com/scientificreports/ dependencies). These different measures were compared and contrasted with three experiments which assessed the consistency of information provided by the FC measures, utility of the FC measures in population-based discrimination and the biological plausibility of using these FC measures.Experiment 1 (E1). The goal of the first experiment was to understand the relative consistency of the different FC measures when evaluated in young and healthy individuals. Within-individual consistency was first established by comparing the overlap (Sørensen-Dice similarity coefficient) of the FC pattern observed in task (motor and verbal) and resting-state functional MRI and then averaging across individuals for each FC measure. Findings from E1: Consistency of functional connectivity. Consistency of all of the identified FCmeasures was evaluated by comparing task functional MRI and resting-state functional MRI in young healthy adults (demographics in Table 1). Since the connectivity pattern may be different for each FC measure, the task functional MRI data offer a form of ground truth as they activate specific regions in the brain which can bear correspondence with those in resting condition. Thresholded FC maps (threshold = one standard deviation above grand mean) between task and resting-state conditions are illustrated for Pearson's correlation in Fig. 2 (subfigure A) and remaining FC measures in Supplementary Fig. 1.Consistency of resting-state FC, quantified by Sørensen-Dice similarity coefficient, is tabulated in Table 2 for the motor and language networks. While mostly comparable across many FC measures, the overlap coefficients for Pearson's correlation are not necessarily the best in any of these tested net...

Functional connectivity of white matter as a biomarker of cognitive decline in Alzheimer’s disease

“…However, this does not indicate WM FC is never able to reflect the change in early stages. First, Pearson's correlation coefficient is the most commonly used metric to measure the FC among GM regions, but its accuracy has been proven to be limited [68]. Given the lower SNR in WM BOLD signals compared to GM signals and the difference in hemodynamic response function between WM and GM [20,69], the Pearson's correlation coefficient appears to be even less accurate to represent WM-GM FC.…”

Parcels and particles: Markov blankets in the brain

“…Note that the (zero lag) correlation is almost zero. This speaks to the potential importance of using cross-covariance functions (or complex cross spectral in frequency space), when characterizing functional connectivity in distributed brain responses (K. J. Friston, Bastos, et al, 2014 ; Mohanty, Sethares, Nair, & Prabhakaran, 2020 ). This brief treatment of extrinsic coupling has made much of the complex nature of dynamical coupling and how it manifests in terms of functional connectivity.…”

Assessment of Flourishing Levels of Individuals by Using Resting State fNIRS with Different Functional Connectivity Measures

“…Also, 7 of 9 classifiers gave higher accuracy results for DTW-ΔHb dataset than for CC-ΔHbO dataset. Compared to CC, DTW is a novel measure in functional connectivity estimation and few studies were performed by utilizing DTW for EEG (Karamzadeh et al, 2013), fMRI (Jin et al, 2020; Linke et al, 2020; Meszlenyi, Hermann, et al, 2017; Mohanty et al, 2020) and fNIRS (Gokcay et al, 2019). In a recent study, it was shown that its efficiency outperformed conventional CC approach to detect atypical connectivity patterns in autism and it was reported that DTW was found sensitive to BOLD signal amplitude which is dependent on Hb concentration (Linke et al, 2020).…”

“…Also, while estimating the connectivity matrices we utilized the conventional Pearson’s correlation coefficient (CC) and Dynamic Time Warping (DTW) distance. DTW is an elastic matching algorithm that allows to capture the lags between two time series and has recently been used as a functional connectivity metric in several neuroimaging studies (Gokcay et al, 2019; Jin et al, 2020; Linke et al, 2020; Meszlényi et al, 2016; Meszlenyi, Buza, et al, 2017; Meszlenyi, Hermann, et al, 2017; Mohanty et al, 2020).…”