Project title: Cross-Domain Data Fusion techniques for vasculitis prediction
Rationale:Fusion of knowledge from diverse datasets with varying dimensionality is challenging in the industrial health sector. We hypothesise that applying novel data fusion techniques derived from other sectors will provide a framework for moving predictive models of relapse towards a commercialisable physician tool, as well as having wider applicability beyond HELICAL.
Objectives:Define external factors (pathogens, weather, pollution) to be fused for predicting flares, employing various methodologies for data fusion (such as stage-based fusion methods, methods that learn new representation of original features extracted from different datasets using deep neural networks and fusion of data based on their semantic meanings) to reach a more generic data fusion methodology (Objective 1) that will be leveraged for relapse prediction machine learning models (Objective 2). In the final year of their project, using algorithms generated by ESRs 1-3, they will create a mock-up physician tool interface prototype (Objective 4, M1.2)
Expected Results:Data fusion techniques that can incorporate diverse data sources; preliminary machine learning algorithms that predict flare