Predictive modelling plays an important role in healthcare, and one that is growing.  We are working at developing improved tools for answering “what if” questions through computational modelling.  Our strategy is to exploit deep networks for this purpose, drawing on their ability to learn complex mappings between inputs (including clinical data) and outputs.  Our lab’s focus in on data that takes the form of a spatial or spatio-temporal nature. The long-term vision is to replace the very long, cumbersome processes that are currently involved in applying numerical modelling to provide patient-specific treatments.  This requires:

  • absorbing numerical models into the predictive capabilities of the mappings performed by deep networks
  • building tools that can explore the space of possible mappings from inputs – such as image data – through to inference or predictions
  • being honest about uncertainty in models and data when making predictions

Some of these goals require “baby steps”, such as  designing the right network architectures and training regimes in order to achieve acceptable performance in real-world data.

Funded Project:  Predicitve Modelling for AF

In collaboration with the Data Science Institute, the Departments of Computing and Aeronautics and the ElectroCardioMaths Group, we have support from the Rosetrees Trustto take the first steps in bringing modelling, machine learning and clinical procedures together to improve the diagnosis and treatment of atrial fibrillation (AF).   This project is an exciting research challenge to better understand and treat AF by:

  • investigating the rich signal content of the intercardiac electrogram and its relationship to disease
  • enhancing the interpretation and fusion of imaging data with other clinical measurements
  • creating a new generation of modelling technologies built on artificial intelligence

To meet this goal, we are exploiting recent developments in machine learning and advanced numerical modelling techniques. At present, we are seeking PhD candidates and post-docs to join this project, starting either in 2017 or 2018.  For further details, see the flyer opportunities in deep learning for predictive modelling. Note that we are also seeking PhD candidates with similar interests, so if you are considering doing a PhD in this area, please check with us.