A Comparative Study of EEG-Based Nonlinear Effective Connectivity for Classifying Alzheimer’s Disease and Frontotemporal Dementia

Abstract

Background: Alzheimer’s disease (AD) and frontotemporal dementia (FTD) are progressive neurodegenerative disorders affecting over 100 million people worldwide. Effective connectivity (EC), which quantifies directional influence between brain regions, offers insight into disrupted neural circuit dynamics. This study aimed to identify neurophysiological biomarkers of AD and FTD and elucidate mechanisms distinguishing them from cognitively normal (CN) controls. Event-related EEG signals from 88 subjects (36 AD, 23 FTD, 29 CN) were analysed. Methods: Node-level graph-theoretical features were derived from three EC measures (phase slope index (PSI), transfer entropy (TE), and a linear state-space model (LSS) inspired by dynamic causal modelling) across five frequency bands. Results: Classification performance was strongest in the delta and beta bands, with hyperlink-induced topic search (HITS) authority and eigenvector centrality emerging as the most discriminative features, respectively. Quadratic discriminant analysis using HITS authority features from LSS in the delta band performed best, achieving 74.1% accuracy, 76.4% precision, 74.1% recall, 74.3% F1-score, and an AUC of 0.82. Conclusion: These preliminary findings suggest that altered nonlinear EC may be associated with AD and FTD pathophysiology, and highlight EEG-based graph-theoretical features as a promising direction for diagnostic decision-support research, pending validation in larger, independent cohorts.