Robust Assessment of EEG Connectivity Patterns in Mild Cognitive Impairment and Alzheimer's Disease

Abstract

The prevalence of dementia, including Alzheimer’s disease (AD), is on the rise globallywith screening and intervention of particular importance and benefit to those with limitedaccess to healthcare. Electroencephalogram (EEG) is an inexpensive, scalable, andportable brain imaging technology that could deliver AD screening to those without localtertiary healthcare infrastructure. We study EEG recordings of subjects with sporadicmild cognitive impairment (MCI) and prodromal familial, early-onset, AD for the sameworking memory tasks using high- and low-density EEG, respectively. A challenge indetecting electrophysiological changes from EEG recordings is that noise and volumeconduction effects are common and disruptive. It is known that the imaginary part ofcoherency (iCOH) can generate functional connectivity networks that mitigate againstvolume conduction, while also erasing true instantaneous activity (zero or π-phase).We aim to expose topological differences in these iCOH connectivity networks using aglobal network measure, eigenvector alignment (EA), shown to be robust to networkalterations that emulate the erasure of connectivities by iCOH. Alignments assessedby EA capture the relationship between a pair of EEG channels from the similarity oftheir connectivity patterns. Significant alignments—from comparison with random nullmodels—are seen to be consistent across frequency ranges (delta, theta, alpha, andbeta) for the working memory tasks, where consistency of iCOH connectivities is alsonoted. For high-density EEG recordings, stark differences in the control and sporadicMCI results are observed with the control group demonstrating far more consistentalignments. Differences between the control and pre-dementia groupings are detectedfor significant correlation and iCOH connectivities, but only EA suggests a notabledifference in network topology when comparing between subjects with sporadicMCI and prodromal familial AD. The consistency of alignments, across frequency ranges, providesa measure of confidence in EA’s detection of topological structure, an important aspectthat marks this approach as a promising direction for developing a reliable test for earlyonset AD.