Identification of epileptic brain states by dynamic functional connectivity analysis of simultaneous EEG-fMRI: a dictionary learning approach

Abstract

Most fMRI studies of the brain’s intrinsic functional connectivity (FC) have assumed that this is static; however, it is now clear that it changes over time. This is particularly relevant in epilepsy, which is characterized by a continuous interchange between epileptic and normal brain states associated with the occurrence of epileptic activity. Interestingly, recurrent states of dynamic FC (dFC) have been found in fMRI data using unsupervised learning techniques, assuming either their sparse or non-sparse combination. Here, we propose an l1-norm regularized dictionary learning (l1-DL) approach for dFC state estimation, which allows an intermediate and flexible degree of sparsity in time, and demonstrate its application in the identification of epilepsy-related dFC states using simultaneous EEG-fMRI data. With this l1-DL approach, we aim to accommodate a potentially varying degree of sparsity upon the interchange between epileptic and non-epileptic dFC states. The simultaneous recording of the EEG is used to extract time courses representative of epileptic activity, which are incorporated into the fMRI dFC state analysis to inform the selection of epilepsy-related dFC states. We found that the proposed l1-DL method performed best at identifying epilepsy-related dFC states, when compared with two alternative methods of extreme sparsity (k-means clustering, maximum; and principal component analysis, minimum), as well as an l0-norm regularization framework (l0-DL), with a fixed amount of temporal sparsity. We further showed that epilepsy-related dFC states provide novel insights into the dynamics of epileptic networks, which go beyond the information provided by more conventional EEG-correlated fMRI analysis, and which were concordant with the clinical profile of each patient. In addition to its application in epilepsy, our study provides a new dFC state identification method of potential relevance for studying brain functional connectivity dynamics in general.

Pushing the Limits of EEG: Estimation of Large-Scale Functional Brain Networks and Their Dynamics Validated by Simultaneous fMRI

“…Although unreported so far, the match between dFC states derived from fMRI and EEG data found in the present study was somewhat expected, considering that a number of studies have already found EEG correlates of dFC fluctuations and brain states measured with simultaneous fMRI (Tagliazucchi and Laufs, 2015), motivated by the yet unclear physiological underpinnings of dFC (Thompson, 2018). These studies were mainly focused on healthy subjects (Chang and Glover, 2010;Allen et al, 2017) and epilepsy patients (Laufs et al, 2014;Lopes et al, 2014;Preti et al, 2014;Omidvarnia et al, 2017;Abreu et al, 2019). Interestingly, when comparing the contrast of interest for mapping the FEPN with the contribution over time of each fMRI and EEG dFC state, we found that the contribution of two matched dFC states based on their spatial correlation were significantly correlated with the FEPN contrast.…”

“…Such window lengths were selected based on a recent meta-analysis revealing that physiologically meaningful, and statistically validated dFC fluctuations can be detected on the fMRI when using a window length between 30 and 60 s (Preti et al, 2017). Additionally, our previous study focused on the detection of epileptic dFC states from simultaneous EEG-fMRI data also showed the ability to detect epileptic dFC states irrespective of the window length within the abovementioned interval (Abreu et al, 2019). With this combination of parameters, comparable properties are expected to be captured on fMRI and EEG-ESI dFC data, while guaranteeing that all points at the end of the dFC data are considered when building the respective last sliding windows.…”

“…In our previous study, we comprehensively compared the performance of several methods (including the most commonly used k-means clustering and principal component analysis) with distinct degrees of sparsity in time, at identifying dFC states associated with epilepsy (Abreu et al, 2019). We found that an l 1 -norm regularized dictionary learning (DL) approach yielded the best results.…”

“…Because of the lack of an independent validation as the one available for the RSN analyses by the template maps, additional statistical tests were then performed here. First, the statistical significance of the dFC states estimated from both fMRI and EEG-ESI data recorded on each run was assessed by building a data-driven null distribution, following the work described in Leonardi et al (2013) and Abreu et al (2019). Specifically, for each run, the fMRI and EEG-ESI C matrices were first FFT transformed and, for each connectivity pair, a random, uniformly distributed, phase was added; the inverse FFT was then applied to transform the data back to the time domain.…”

“…dFC correlates of disease are believed to be particularly relevant for its characterization, and recent studies have demonstrated that dFC discriminates healthy from patient populations better than static FC (Du et al, 2018). Under the assumption that brain function dynamics can be described by a limited number of states (Preti et al, 2017), a significant number of studies have dedicated to the identification of such brain states from the dFC, by applying pattern recognition techniques, particularly clustering (Allen et al, 2014), principal component analysis (Leonardi et al, 2013), and dictionary learning Abreu et al, 2019).…”