Supervisor

Suitable for

Abstract

Supervisors: Hang Dong (University of Exeter) and Michael Benedikt (Oxford)

Early dementia prediction is an urgent problem. The goal is to improve the clinical pathway for individuals who are at risk of dementia due to well-established risks such as cognitive decline, lifestyle, medical or genetic factors. Artificial Intelligence has the potential to offer a route to personalisation of risk profiling to detect early, mild cognitive impairment cases.

There are two key challenges in early dementia modelling, which are at the core of this project:

(1)          The linking and generative modelling of longitudinal, tabular, cognitive test data with external knowledge bases, and

(2)          the meaningful explanation of the prediction, using linkage to external knowledge bases within explanation.

The project will build on recent advances in prediction methods [1] for tabular data understanding e.g., using tables in csv or excel format that record information from participants across several timepoints, including clinical data and answers to questions.

We will extend these methods with linkages to domain-specific knowledge bases [2] (e.g., constructed from the literature on cognitive tests and dementia risks with the tabular data). This approach is expected to enhance both prediction performance and interpretability, providing more robust insights into dementia progression and associated risk factors. The project will mainly work on the data from the PROTECT platform [3], containing a cohort of 30,000 older adults in community and primary care settings with ten-year longitudinal data.

We will explore enhancing explainability through connections within the tabular data and between the tabular data and external sources. For example, we can link datatypes in the table to knowledge bases derived from the literature, and these can be useful in providing more interpretable predictions [5].

The project team will work with medical experts at University of Exeter, both in providing background and in assessment.

The project requires good development skills and a good general background in machine learning. No specialised background on dementia or tabular data learning is required.

 [1] Shmatko A, Jung AW, Gaurav K, Brunak S, Mortensen LH, Birney E, et al. Learning the natural history of human disease with generative transformers. Nature. 2025 Sept 17;1-9. doi: 10.1038/s41586-025-09529-3

[2] Chen J, Dong H, Hastings J, Jiménez-Ruiz E, López V, Monnin P, Pesquita C, Škoda P, Tamma V. Knowledge Graphs for the Life Sciences: Recent Developments, Challenges and Opportunities. Transactions on Graph Data and Knowledge. 2023 Dec 19;1(1):5-1. doi: 10.4230/TGDK.1.1.5

[3] University of Exeter. PROTECT Study: Research to support healthy ageing and reduce dementia risk. https://www.exeter.ac.uk/research/dementia-research/research/protect/#a0. Accessed 28 Sep 2025. (Led by Prof Anne Corbett and the team)

[4] Hotchkiss L, Squires E, Gallacher JE, Newbury M, Morris C, Lyons RA, et al. Enabling Advanced Multi-Modal Neuroimaging Analysis within a Trusted Research Environment. medRxiv; 2024 Feb 13. doi: 10.1101/2024.02.13.24302751. (See datasets at https://portal.dementiasplatform.uk/data/imaging-matrix/)

[5]  Longo L, Brcic M, Cabitza F, Choi J, Confalonieri R, Ser JD, et al. Explainable Artificial Intelligence (XAI) 2.0: A manifesto of open challenges and interdisciplinary research directions. Information Fusion. 2024 June 1;106:102301. doi: 10.1016/j.inffus.2024.102301