Author Identifier
Heng Zhang: http://orcid.org/0000-0002-4358-3635
Date of Award
2026
Keywords
Alzheimer's disease, mild cognitive impairment, EEG–fMRI coupling, functional neuroimaging, deep learning, multimodal biomarkers
Document Type
Thesis
Publisher
Edith Cowan University
Degree Name
Doctor of Philosophy
School
School of Medical and Health Sciences
First Supervisor
Simon Laws
Second Supervisor
Shuhua Ma
Third Supervisor
Wei Wang
Fourth Supervisor
Lei Xie
Abstract
Background: Alzheimer's disease (AD) is a leading cause of dementia, accounting for 60–70% of all cases worldwide, and poses a growing burden on healthcare systems and societies. While early detection during the prodromal stage of mild cognitive impairment (MCI) is critical for timely intervention, current diagnostic methods remain limited in capturing the complex neural alterations underlying cognitive decline. Functional neuroimaging techniques such as fMRI (Functional Magnetic Resonance Imaging) and EEG (Electroencephalography) offer complementary insights into brain dynamics, yet their isolated use fails to resolve the whole temporal-spatial interplay of AD-related dysfunction. Simultaneous EEG–fMRI, combined with advanced deep learning, presents a promising integrative approach to bridge this gap.
Aims: This thesis aimed to (1) characterize frequency-dependent alterations in brain function using resting-state fMRI and EEG; (2) quantify and interpret cross-modal EEG–fMRI coupling at static and dynamic timescales; (3) investigate the relationship between coupling metrics and cognitive performance; and (4) develop an interpretable deep learning framework that integrates multimodal neuroimaging features to improve diagnostic classification and uncover disease-related biomarkers.
Methods: Simultaneous resting-state EEG and fMRI data were acquired from 99 participants (14 AD, 45 MCI, 40 healthy controls). Preprocessing was conducted using standardized pipelines. Functional brain alterations were examined across frequency bands using fMRI metrics and EEG-based network analyses. Cross-modal coupling was assessed via general linear modelling and sliding-window correlation. Cognitive correlates (MoCA, MMSE) were analyzed. Deep learning models included unimodal EEG and fMRI architectures and a multimodal fusion network with cross-modal attention, evaluated using stratified cross-validation.
Results: This thesis revealed robust spatial and temporal neural feature alterations across the AD continuum. Frequency-specific disruptions in fMRI metrics were observed in default mode and salience networks, with Slow-5 ReHo and Slow-4 ALFF showing enhanced sensitivity to early-stage changes (Chapter 3). Whilst, EEG analyses showed progressive spectral slowing, breakdown of in-phase synchronization, diminished global efficiency, and disrupted modular architecture (Chapter 4). Further, cross-modal EEG–fMRI coupling strength increased with disease severity and demonstrated significant spatial heterogeneity, particularly in frontotemporal and midline regions. Notably, coupling metrics were significantly associated with cognitive scores, highlighting their potential as integrative biomarkers (Chapter 5). Finally, the deep learning framework, implemented to integrate EEG and fMRI, outperformed unimodal baselines, achieving up to 80% classification accuracy. Resultant, attention maps and coupling-informed features provided interpretable links between network dysfunction and clinical phenotype (Chapter 6).
Conclusions: This thesis presents a unified multimodal framework for characterizing neural alterations in Alzheimer's disease, combining simultaneous EEG–fMRI with explainable deep learning. The findings underscore the synergistic value of spatial-temporal integration for early detection and mechanistic interpretation of cognitive decline. The proposed approach offers a scalable and biologically grounded pathway toward precision diagnostics in neurodegenerative disorders.
Access Note
Access to chapters 3-7 of this thesis are not available
Access to this thesis is embargoed until 25th April 2031
Recommended Citation
Zhang, H. (2026). Multimodal integration of simultaneous EEG–fMRI and deep learning for early detection and mechanistic characterization of Alzheimer's disease. Edith Cowan University. https://doi.org/10.25958/rv7a-rj36