Alzheimer's disease (AD), the most common form of dementia, is an incurable neurodegenerative disease that affects millions of elderly people worldwide. Detecting the disease in its early stages is the key for a more effective treatment. AD is a multifactorial disease, where several biomarkers represent different pathophysiological processes in the brain, with distinct progression paths over time. Methods to facilitate the integration and interpretation of longitudinal, heterogeneous medical data could be of benefit for a better understanding of the disease and its progression. In this thesis, we present statistical and machine learning methods and studies for early detection and to assess disease progression.
Contributions of this thesis are as follows:
First, we present a review on machine learning applications in AD using longitudinal neuroimaging data: we analyze their approach to typical challenges in longitudinal data analysis and show that machine learning methods using this type of data have potential to improve disease progression modelling and computer-aided diagnosis.
Our second contribution is a study of AD subtyping using novel plasma-based blood biomarkers. We used a multivariate, unsupervised multiple kernel learning method over blood-based biomarkers to find subgroups of patients defined by distinctive blood biomarker profiles, and we analyze those subgroups using cross-sectional and longitudinal neuroimaging data.
Our third contribution is a novel method based on recurrent, multimodal variational autoencoders to model the progression of the disease. It can use a variable number of modalities and time-points across different subjects, and we show its performance quantitatively and qualitatively.
Our fourth and final contribution is an analysis of the impact of APOE e4 gene dose and its association with age on hippocampal shape, assessed with multivariate surface analysis, using a cognitive healthy, e4-enriched cohort.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados