Estarellas Garcia, Maria del Mar; (2023) Multimodal modelling of early neurodegeneration and Alzheimer's disease progression. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Thesis_MarEstarellasGarcia.pdf - Other Download (13MB) | Preview

Abstract

Alzheimer’s disease (AD) is a heterogeneous condition with a long pre-symptomatic phase. Treatments are likely to be more effective if administered at earlier stages in targeted groups. However, early detection and stratification is challenging, requiring the incorporation of biomarkers that are sensitive to pre-symptomatic disease, and the disentanglement of the different factors that contribute to disease heterogeneity. Two key contributors to AD heterogeneity are temporal heterogeneity (individuals are at a range of underlying stages) and subtype heterogeneity (individuals belong to different subgroups with distinct temporal progression patterns). Whilst a range of clustering methods have been used to characterise AD subtypes, these methods fail to account for temporal heterogeneity, potentially conflating disease subtypes with disease stages and precluding subtyping of individuals at early stages. SuStaIn (Subtype and Stage Inference) extracts both the phenotypical and temporal heterogeneity of AD by using a data-driven combination of clustering and disease progression modelling. However, SuStaIn does not currently account for missing data, often limiting the application of SuStaIn to one type of biomarker and thus excluding key information about the diverse set of early changes that occur in AD. In this thesis, I develop a missing data adaptation of SuStaIn (“missing-data SuStaIn”) and benchmark it against other missing data algorithms. I apply missing-data SuStaIn to a research cohort (ADNI) to characterise the heterogeneity of multi-modal biomarker trajectories in an amyloid-enriched cohort, demonstrating that using multi-modal data significantly improves the prediction of conversion between diagnoses; and unravels a higher number of disease subtypes than when using MRI only. I then apply missing-data SuStaIn to a population cohort (Insight46) to study the heterogeneity of multi-modal biomarker trajectories in an elderly population, uncovering three subgroups with early biomarker changes resembling pre-symptomatic Alzheimer’s disease. The missing-data adaptation to SuStaIn developed in this thesis has wide potential applications to other diseases and developmental processes.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as JSONCSVXML

Archive Staff Only

View Item