Binarized Brain Connectivity: A Novel Autocorrelation-Based Iterative Synchronization Technique

摘要

The brain is a highly interconnected neurobiological system. Network-level characterization is thus largely performed to understand brain functioning. Brain activity can be captured through modalities like functional magnetic resonance imaging (fMRI) and electroencephalogram (EEG). Brain networks are then estimated through pairwise relationships between brain regions using brain connectivity. Functional connectivity, which measures the degree of coactivation between two brain regions, is estimated from pairwise EEG (or fMRI) time series using a measure of synchronization like Pearson's correlation. However, all such measures suffer from the fact that they are continuous variables, with values often lying in the ambiguous range (say 0.3-0.7) wherein it is difficult to infer whether the two time series are actually synchronized or not. This makes the interpretation of findings challenging. Synchronization measures are also largely corrupted by noise. In this paper, a novel autocorrelation-based iterative synchronization (ABIS) technique is proposed, which provides binary synchronization values (0 = not synchronized, 1 = synchronized). It is entirely data driven with no assumptions, input parameters, or arbitrary choices. We demonstrate that ABIS resolves ambiguous synchronizations and provides reliable, robust, and neurobiologically meaningful binary synchronization values. ABIS also performs better than conventional synchronization on all these faculties. This technique has tremendous applications in brain functional connectivity analysis. Complex network modeling of the brain using graph theoretic techniques largely require binary connectivity matrices, which are often obtained by arbitrarily thresholding continuous connectivity matrices. Such practice usually sophisticates the analysis or yields unreliable results. The use of ABIS could entirely eliminate these issues since it provides binary connectivity matrices in a single step, without assumptions or arbitrary choices. Additionally, it resolves ambiguous connectivities to provide a set of sure-connections and sure no-connections, which improves the interpretability of connectivity results and enhances noise robustness (which plagues connectivity analysis). This study has potential applications in network modeling of the brain, and graph-theoretic analysis in general.