Approved research

Building a multimodal imaging classifier for prediction of clinical outcomes in brain related disorders

Lay summary

The aim of this project is to improve the prediction of disease health status using brain imaging data. We benefit from Machine Learning techniques to develop a multimodal classifier working on features extracted from different types of Brain MRI (including anatomical, diffusion and functional). Our primary goal is to build a classifier to predict a phenotype of interest, such as a type of brain disorder or disease prognosis, from multiple imaging modalities. We also aim to make an emphasis on learning interpretable features, allowing to find associations between image patterns and phenotype. Understanding the complexity of our brain is a key factor to facilitate clinical prediction, prevention and treatment of brain disorders, such as Dementia, Alzheimer's disease, and Unipolar Depression. While it is known that certain biomarkers extracted from specific imaging modalities might be informative for a given disease, we are still far from understanding how we could benefit from efficiently combining multiple imaging modalities and other sources of information. The UK Biobank is an extremely valuable data source for this purpose, providing imaging data obtained with different techniques as well as clinical information on a large scale number of subjects. We will build a machine learning pipeline to perform the following two key steps: (1) learn informative features across multiple imaging modalities and (2) use those features to design improved clinical predictors. Building on state-of-the-art machine learning methods, we will develop an algorithm to predict health-related phenotypes from multiple Brain MRI data and to identify the features that are most associated with a disease and its progression. Upon completion, our research will help clinicians make better use of machine learning to improve diagnosis and personalise medical treatment. Full cohort for brain imaging data and the complementary clinical information.