University of Leeds Featured PhD Programmes
Catalysis Hub Featured PhD Programmes
University of Glasgow Featured PhD Programmes
University of Kent Featured PhD Programmes
University of Reading Featured PhD Programmes

MRC DTP 4 Year PhD Programme: AI-Spot-Dementia (AISDA): developing machine learning algorithms to predict the risk of dementia for patients with Type 2 Diabetes

This project is no longer listed on and may not be available.

Click here to search for PhD studentship opportunities
  • Full or part time
    Dr J Zhang
    Prof D Steele
    Dr A Doney
    Prof E Trucco
  • Application Deadline
    No more applications being accepted
  • Competition Funded PhD Project (European/UK Students Only)
    Competition Funded PhD Project (European/UK Students Only)

About This PhD Project

Project Description

The aim of this PhD proposal is to develop novel machine learning algorithms (e.g., deep learning) to predict the risk of dementia in Type 2 diabetes, and detect early signs of such disease with association of clinical and genomic data in clinical settings. There are strong links between Type 2 Diabetes (T2D) and dementia. With increasing numbers of people developing T2D, detecting early signs of dementia is important to better understand how it can be delayed or prevented. Brain scans contain large amounts of information about brain health not used routinely by doctors, because it is difficult to measure. We hypothesize that the advance of machine learning could provide a way to better diagnosis and predictions of disease risk. We have routinely acquired NHS brain scans in the large GoDARTS study population, which provides a solid platform for us to develop and test our algorithms.

GoDARTS comprises over 18,000 individuals who provided a sample of blood for genetic studies and consented for this information to be linked their electronic medical records, including routinely acquired clinical brain scans. There are currently over 1,400 incident cases of dementia in GoDARTS, many adjudicated by a nurse-led validation process. In our £1.1M EPSRC-supported collaboration on retinal biomarkers for dementia, we have successfully developed the technical and governance infrastructure in HIC to link retinal images from the Diabetes Retinal Screening service within a Safe Haven environment, to analyse retinal microvascular determinants of neurodegeneration. This framework has resulted in approximately 800 MRI brain scans of GoDARTS participants, acquired as part of usual NHS care, being anonymised and made available for research in the Safe Haven. It is anticipated that, during the work proposed, all historical brain scans from GoDARTS participants will be made available through parallel initiatives.

Brains scans will be anatomically normalised and intensity standardised as described elsewhere. Fully automated ML algorithms comprising shallow methods based on support vector machine (SVM) and deep end-to-end convolutional neural network (CNN) models will then be developed to predict the risk of dementia. The work includes: i) Low-level handcrafted feature extraction (e.g. WM/GM probabilities from SPM spatial normalisation and MR8 features which capture vascular structures) creating a pool of potential biomarkers. Variants of SVMs will be developed as baseline for classification and biomarkers identification. ii) Building a hybrid CNN model motivated by heterogeneity in the data with learning driven mainly by brain scans, but also by cross-linked data (e.g. blood pressure, APOE4 status, VAMPIRE retinal markers), embedded in one of its fully connected layer. This approach has a potential to discover hypothesis-free features with better prediction performance. Automated brain age prediction will also be explored using existing public dataset on healthy brain using a state-of-the-art deep CNN serving as an input to the model for predicting the dementia risk for individual patients. Prior work of the team will serve as the starting point.


Hongwei Li, Gongfa Jiang, Jianguo Zhang, Ruixuan Wang, Zhaolei Wang, Wei-Shi Zheng, Bjoern Menze, Fully Convolutional Network Ensembles for White Matter Hyperintensities Segmentation in MR Images. NeuroImage, to appear, 2018 (Winning method for the International challenge on brain White Matter Hyperintensities (WMH) segmentation, MICCAI 2017).

Hébert, HL, Shepherd, B, Milburn, K, Veluchamy, A, Meng, W, Carr, F, Donnelly, LA, Tavendale, R, Leese, G, Colhoun, HM, Dow, E, Morris, AD, Doney, AS, Lang, CC, Pearson, ER, Smith, BH & Palmer: Cohort Profile: Genetics of Diabetes Audit and Research in Tayside Scotland (Go DARTS). International Journal of Epidemiology, vol. 47, no. 2, pp. 380-381j. DOI: 10.1093/ije/dyx140

Pellegrini, E., Ballerini, L., Hernandez, M. D. C. V., Chappell, F. M., González-Castro, V., Anblagan, D., ... Wardlaw, J. M. (2018). Machine learning of neuroimaging for assisted diagnosis of cognitive impairment and dementia: A systematic review. Alzheimer’s and Dementia: Diagnosis, Assessment and Disease Monitoring, vol 10, 519-535. DOI: 10.1016/j.dadm.2018.07.004.

FindAPhD. Copyright 2005-2019
All rights reserved.