Artificial networks for predicting lifelong brain health

The project aims to develop Artificial intelligence methods for application to mental health. Capitalising on existing data sets to develop friendly and cost-effective diagnostic screening that is scalable to clinical practice across the Health Sector, the propped project is at the core of Data-Driven economy and the ‘Data to early diagnostics and precision medicine’ programme.Recent advances in brain imaging technology provide means for identifying biomarkers, such as tau- or amyloid-PET imaging, that are highly prognostic of neurodegenerative decline for dementia. However, these measurements are invasive and cannot be included in routine clinical practice due to resource constraints. Further, drug discovery focuses on animal models using invasive techniques (e.g. microscopy) that are not applicable to humans. Thus, there is pressing need for in silico cost-effective tools that extract highly predictive markers from patient data and can drive drug discovery and precision treatment. We will build robust models using metric learning algorithms on large-scale datasets that incorporate invasive measurements (e.g. PET imaging that requires radioactive tracers) as privileged information. We will then use these models to predict health vs. disease using independent data sets that contain data that is less predictive; that is, data typically collected in routine clinical practice (e.g. memory screening tests, structural MRI scans).

This project is in collaboration with The Centre for Mathematical Imaging in Healthcare (CMIH) and AstraZeneca’s Quantitative Biology team