Escudero Rodriguez , Javier

Thompson, Jack

Arslan, Tughrul

Campbell-Cousins, Avalon

2026-05-19T14:06:04Z

2026-05-19

Neurodegenerative diseases, most prevalently Alzheimer’s disease (AD), cause progressive cognitive and intellectual decline, placing immense emotional and financial strain on families and government services. Primary methods in the diagnosis of AD typically focus on cognitive questionnaires, screening tests, medical history, and neuropsychological examinations. However, early detection of AD remains difficult, with gaps in our understanding of how early stages contribute to the disease’s progression. To this end, a range of neuroimaging methods both structural, such as structural magnetic resonance imaging (sMRI) and diffusion tensor imaging (DTI), and functional, including functional magnetic resonance imaging (fMRI) and electroencephalography (EEG), have been used to study changes in the brain across the AD continuum and uncover biomarkers of disease progression. These advancements in brain imaging have facilitated a network-based model of the brain where brain regions are linked according to anatomical connections or functional associations. This model captures the inter-dependencies and interactions between brain regions that drive complex cognitive processes. These brain regional interactions can be modelled as single-layer networks or as a collection of brain networks (multiplex networks) which evolve across time or other indexed data. Brain networks organize and reorganize themselves in a multitude of ways such as in response to task stimuli or as a result of damage due to disease. Organizational topological structures, captured as modules (groups) composed of a collection of brain regions, can be dynamically explored and are of particular interest for diseases such as AD where progression causes abnormal altering of these modular structures. However, modularity in its current state is limited in its use as a metric describing the total extent to which the brain is segregated into distinct modules. It does not currently describe the individual contributions of individual brain regions to overall brain modularity. To tackle this, we introduce nodal modularity (nQ) as a novel graph measure that extends classical modularity to individual nodes for both single-layer and multiplex networks. We assess the novelty of nQ against other common single-layer and multiplex measures of node influence. Additionally, we explore the hypothesis that nQ would yield novel insights into the progression of amnestic Mild Cognitive Impairment (aMCI, a prodromal stage of AD) given that global changes in modularity have been previously observed along this continuum. This is investigated in single and multiplex networks of a visual short-term memory binding task (VSTMBT – a cognitive biomarker of AD) constructed from DTI and task-fMRI data. The results indicated that nQ could effectively characterize the transition point from MCI to AD (MCI converters, or MCIc). Additionally, results corroborated previously understood progression pathways of AD including the identification of key subnetworks affected by the disease (visual, limbic and paralimbic), along with agreement with amyloid-β and tau deposition for subjects with poor visual short-term memory binding. Following the introduction of nQ, we expand the study to larger multiplex networks with a focus on covering typical constructions in neuroimaging research. We verify that nQ is distinct from other multiplex measures of node influence across a number of multiplex surrogate networks. We find that, as network size and the number of network layers increase, that nQ captures unique information distinct from other multiplex network metrics, reflecting nQ’s ability to capture more wide-reaching changes in network topology than the typical local measures used (multiplex clustering coefficient, multiplex PageRank, and degree). Additionally, we explore specific cases where nQ does not provide a significant advantage as a network measure over the state of the art, largely driven by cases of highly modular networks with high within module connectivity and low between module connectivity. Furthermore, we propose collapsibility as a novel method of simplifying analyses of nQ in larger networks, explored in tandem with multiplex flexibility. We apply these to multiplex frequency-based networks of source-space EEG data during the VSTMBT, and analyse changes along the healthy to MCI progression of AD. We found changes in nQ and collapsibility which agree with known changes in grey matter and with abnormal functional connectivity around the thalamus as a result of MCI and AD. Lastly, we explore how nQ can facilitate the simulation of AD progression by developing novel targeted attack models. Specifically, we investigate the current state of the art in targeted attacks in our task-fMRI data in the modelling of healthy to MCIc progression. We find that prior research in targeted attacks that focus on node degree were not sufficient in describing this progression from healthy to MCIc for task-fMRI. To tackle this, we propose a brain lobebased targeting system informed by prior findings on nQ progression and established hypotheses on network damage in AD. Additionally, we confirm that improvements in simulated progression using the lobe-based method are not explainable solely by preferential attacks on long distance connections, achieved through the introduction of a targeted attack model based on Euclidean distance. We then leverage nQ to improve the lobe-based model and show that both lobe-based and nQ-based targeted attack models improve on the state of the art. In sum, this thesis highlights the importance of considering granular changes in network topology to capture complexities in brain structure and function. Specifically, through the establishment of a novel metric of granular community structure, nQ, providing methodological considerations for the use of this measure in larger networks, and in exploring the simulation of granular changes due to AD through novel targeted attack models. Given modularity’s widespread use as a global measure, nQ represents a significant advancement, providing a granular measure of network organization applicable across disciplines. In regard to AD, this thesis motivates additional study of nQ in characterizing the stages of MCI and particularly in disentangling MCI from MCI converters during the VSTMBT.

https://era.ed.ac.uk/handle/1842/44719

https://doi.org/10.7488/era/7234

en

Campbell-Cousins A, Guazzo F, Bastin ME, Parra MA, Escudero J. Multiplex nodal modularity: A novel network metric for the regional analysis of amnestic mild cognitive impairment during a working memory binding task. PLOS ONE. 2025;20(8):e0328736. doi:10.1371/JOURNAL.PONE.0328736

Fabila-Carrasco JS, Campbell-Cousins A, Parra-Rodriguez MA, Escudero J. Graph- Based Permutation Patterns for the Analysis of Task-Related FMRI Signals on DTI Networks in Mild Cognitive Impairment. ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings. 2024; p. 2076–2080. doi:10.1109/ICASSP48485.2024.10447332

Roy O, Campbell-Cousins A, Carrasco-Stewart J, Parra MA, Escudero J. Activated Permutation Entropy for Graph Signals. Research Square (preprint). 2025;doi:10.21203/RS.3.RS-7220016/V1

Alzheimer’s disease

amnestic mild cognitive impairment

aMCI

MCI converters

early detection

VSTMBT

nodal modularity

brain networks

diffusion tensor imaging

combined network layers

Granular changes in brain networks: multiplex nodal modularity for Alzheimer's disease characterization and simulation

Thesis

Doctoral

PhD Doctor of Philosophy