Network-level permutation entropy of resting-state MEG recordings: a novel biomarker for early-stage Alzheimer’s disease?

Abstract

Abstract Objective. Increasing evidence suggests that measures of signal variability and complexity could present promising biomarkers for Alzheimer’s disease (AD). Earlier studies have however been limited to the characterization of local activity. Here, we investigate whether a network version of permutation entropy could serve as a novel biomarker for early-stage AD. Methods. Resting-state source-space magnetoencephalography was recorded in 18 subjects with subjective cognitive decline (‘SCD’) and 18 subjects with mild cognitive impairment (‘MCI’). Local activity was characterized by permutation entropy (PE). Network interactions were studied using the inverted Joint Permutation Entropy (JPEinv), corrected for volume conduction. Results. The JPEinv showed a reduction of nonlinear connectivity in MCI subjects in the theta and alpha band. Local PE showed increased theta-band entropy. Between-group differences were widespread across brain regions. ROC analysis of classification of MCI versus SCD subjects revealed that a linear regression model trained on JPEinv features (78.4% [62.5–93.3%]) slightly outperformed PE (76.9% [60.3–93.4%]) and relative theta power based models (76.9% [60.4–93.3%]). Conclusion. Classification performance of theta JPEinv was at least as good as the relative theta power benchmark. The JPEinv is therefore a potential biomarker for early-stage AD, and should be explored in larger studies.