The research objectives are to (1) construct a quantitative etiology model of mixed Alzheimer’s disease (AD) and vascular dementia (VaD), and (2) develop the necessary underlying quantitative T1-weighted (T1-w) and/or T2-weighted (T2-w) magnetic resonance imaging (MRI) imaging biomarkers relating to white matter and vascular pathologies.
Quantitative etiology model of mixed AD and VaD
The data-driven quantitative modelling of disease development (i.e., the quantitative etiology model) will be constructed as maximum-likelihood or penalized maximum-likelihood models of development of biomarkers. Development will be modelled as a function of a continuum variable of staging or “disease-age” that is to be inferred for the individual. The result is a set of parametric curves of biomarker values and their variation as a function disease development.
Deep learning-based white matter QIB
White matter pathologies are modelled and their quantification will be based on volumetric MRI T1-w and T2-w scans. Normally, white matter hyper intensities would be derived from fluid-attenuation inversion recovery (FLAIR) whereas white matter changes such as microstructural changes relating to phenomena such as demyelination, changes in fibre densities, and potential build-up of plaque would be measured as changes in diffusivity using diffusion weighted MRI (DWI). The objective is to research the potential to derive informative measures of pathology load and location from T1-w and T2-w imaging only. Other modalities (FLAIR/DWI) and non-imaging biomarkers will be used for training using deep learning.
Deep learning-based vascular QIB
Micro bleeds are another type of vascular pathology. Unlike the ischemic lesions that appear as white matter hyper intensities, micro bleeds are typically observed on T2* sequences. The objective of the vascular pathology research is to use T2* images in conjunction with T1-w and T2-w to learn the potential change of appearance of vascular pathologies in T1-w and/or T2-w, and to create a quantitative summary of this appearance as a quantitative biomarker using deep learning.