From behavioural patterns to rain barriers: representation learning for ageing and dementia
File(s)
Author(s)
Fletcher-Lloyd, Nan
Type
Thesis
Abstract
The global population is ageing, placing a significant strain on social and healthcare services. Age is also the most significant risk factor for dementia. With a large proportion of long-term care for older adults focused on supporting people living with dementia, developing more efficient approaches to support clinical discovery and improve care is a top priority. Biomedical data contains a wealth of information that can be leveraged for a range of clinical applications, from drug discovery to diagnosis to care delivery. However, analysis and annotation of such data is challenging due to sheer volume and complexity. Representation learning can automate this process, uncovering informative features to facilitate the development of scalable and generalisable models that offer clinically meaningful insights into disease mechanisms, and for clinical decision-making.
The research presented in this thesis focuses on developing representation learning methods for a range of tasks and evaluating these approaches on real-world data. This work used a variety of data modalities, from remote monitoring time series data to high-resolution electron microscopy image data, to inspire three main studies: the development of a proof-of-concept Markov model for patient-specific anomaly detection in daily activity patterns; the design and validation of an interpretable predictive approach for assessing cognitive decline using in-home sleep monitoring; and the development of a deep transfer learning framework for data-driven analysis of EM images to identify the architecture of the blood-brain barrier, a structure crucial to maintaining brain homeostasis, in aged versus young mouse brains.
This research contributes novel representation learning systems to improve biomedical research from biological discovery to clinical monitoring, with the potential to support both researchers and clinicians in furthering our understanding of disease mechanisms and developing more targeted clinical interventions to improve quality of life for older adults and people living with dementia.
The research presented in this thesis focuses on developing representation learning methods for a range of tasks and evaluating these approaches on real-world data. This work used a variety of data modalities, from remote monitoring time series data to high-resolution electron microscopy image data, to inspire three main studies: the development of a proof-of-concept Markov model for patient-specific anomaly detection in daily activity patterns; the design and validation of an interpretable predictive approach for assessing cognitive decline using in-home sleep monitoring; and the development of a deep transfer learning framework for data-driven analysis of EM images to identify the architecture of the blood-brain barrier, a structure crucial to maintaining brain homeostasis, in aged versus young mouse brains.
This research contributes novel representation learning systems to improve biomedical research from biological discovery to clinical monitoring, with the potential to support both researchers and clinicians in furthering our understanding of disease mechanisms and developing more targeted clinical interventions to improve quality of life for older adults and people living with dementia.
Version
Open Access
Date Issued
2025-09-30
Date Awarded
2026-02-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Barnaghi, Payam
Sponsor
The UK Dementia Research Institute (UK DRI) Care Research and Technology Centre (CRT), funded by the Medical Research Council (MRC), Alzheimer’s Research UK, and Alzheimer’s Society
The UK Research and Innovation (UKRI) Engineering and Physical Sciences Research Council (EPSRC) PROTECT Project
Grant Number
UKDRI-7002
EP/W031892/1
Publisher Department
Department of Brain Sciences
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)