Machine Learning for Early Diagnosis of Dementia

Combining Supervised and Unsupervised Machine Learning for Early Diagnosis of Dementia

Despite the increasing availability of health data, clinically translatable methods to predict the conversion from Mild Cognitive Impairment (MCI) to dementia are still lacking. MCI represents a precursor to dementia for many individuals; however, some forms of MCI tend to remain stable over time and do not progress to dementia. In fact, only a fraction of 20–40% of MCI individuals converts to dementia within 3 years following the initial diagnosis. In order to identify personalized strategies to prevent or slow the progression of dementia and to support the clinical development of novel treatments, we need to develop new approaches for modelling disease progression that are able to differentiate between progressive and non-progressive MCI subjects.

This talk will cover the development of a novel hybrid prognostic framework utilising longitudinal information encoded in efficient, cost-effective, and non-invasive markers to identify subjects with MCI that are at risk for developing dementia. The main novelty of this ML framework lies in using the output of the rigorously designed unsupervised learning approach as input in supervised ML models. This is the first time that such an approach has been utilized for dementia prognosis. Our results show that exploiting information generated through unsupervised learning can deliver stronger predictions in a supervised learning task.

Bio:

Dr Magda Bucholc is a Lecturer in Data Analytics and George Moore Fellow at the Ulster University School of Computing, Engineering, and Intelligent Systems.

Magda works on the development and implementation of advanced analytics and machine learning methodologies in clinical decision making, in particular, to improve current disease risk prediction models, enable earlier diagnosis, optimize allocation of health care resources, and deliver more tailored treatments to patients. Her current research interests include statistical modelling, machine learning, pattern recognition, Bayesian methods, and causal inference. Magda is the Early Career Co-Lead for the Deep Dementia Phenotyping Network. She also acts as the Northern Ireland representative on the British Standards Institution committee, developing a standard for safe and effective use of AI in health and social care. Her research has been supported by funding from ESRC, ARUK, INTERREG VA, Department for the Economy, Science Foundation Ireland, HSC R&D Public Health Agency, Global Challenge Research Fund, and Ulster University Research Challenge Fund.