Machine Learning in the Wild for Cognitive Disorders

: Investigating Dementia Progression Through Continuous Monitoring in Smart Homes

Abstract

The need to automatically monitor the state and progression of chronic neurological diseases
such as dementia, together with the emergence of state-of-the-art sensing platforms for the
home environment, offer unprecedented opportunities for automatic behavioural monitoring
as a proxy of the disease state. However, when these platforms have been deployed, data challenges, including the lack of reliable annotations, data quality and availability, limit the applicability of standard machine learning techniques. Under these challenges, this thesis specifically seeks to characterise behavioural signatures of mild cognitive impairment (MCI) and dementia in smart homes with a novel dataset of longitudinal sensor data and with only partial annotations.
This thesis was undertaken within the ContinUous behavioural Biomarkers Of cognitive Impairment (CUBOId) project that collected data from participants in the UK for up to two years. To
complement existing tools on the assessment of data quality, and inspired by the nature of our data, we first present hypothesis tests to check whether a given dataset of instance-label pairs has been corrupted with label noise. We also introduce the bespoke machine learning pipeline devised for the CUBOId project specifically that accounts for partial annotations to produce behavioural metrics of key symptoms of the disease. Finally, by exploring the longitudinal sensor data, we investigate the behavioural patterns derived from the machine learning pipeline, the proxies of disease status, and their correlations with the clinical evaluation of the project’s participants.

Collectively, this work introduces a robust pipeline for the analysis and understanding of relevant
symptoms of MCI and dementia. We show that the progression of MCI or dementia can be identified
by an increase in these frequently observed symptoms. This study serves as one of the first steps
towards linking machine learning techniques and dementia research under free-living conditions
and offers a complementary view of these diseases.

Date of Award	13 May 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Raul Santos-Rodriguez (Supervisor) & Niall Twomey (Supervisor)