AI: from high energy physics to medical applications

Student: Robert McNulty
Supervisors: Nikos Rompotis (UoL), Monica D’Onofrio (UoL), Dennis Kehoe (AIMES)
Institution: University of Liverpool

Artificial intelligence techniques are more and more used in many fields. The Liverpool team working on the ATLAS experiment at the Large Hadron Collider (LHC) has been involved for many years in the development and performance evaluation of neural networks applied to high energy physics data analysis. AIMES – a UK owned company that provides cloud solutions and services – has developed extensive expertise on health data research and use automated methods based on AI techniques for healthcare in applied medical science.

The student will work with particle physicists and AIMES experts in a dynamic project, having access to postgraduate training in particle physics and to the LIV.INNO structured training. During their Year 1 of PhD, they will collaborate with ATLAS members at Liverpool in the search for Higgs-pair production at the ATLAS experiment using data from the Run 3 of the LHC. They will commission and improve ML algorithms used for efficient identification of objects relevant for the analysis, including tau-lepton and heavy-flavour jets. These techniques will be then applied in the future Higgs-pair analysis to improve its sensitivity.

For the following two years (Year 2 and 3), the student will work alongside experts at AIMES to evaluate the performance of convolutional neural networks to magnetic resonance (MR) images data. AIMES, in collaboration with Barts Hospital in London, has access to a large database of MR images and has already developed Fully Convolutional Networks (FCNs) that can analyse them. Their work will ultimately target the development of a validation methodology for these networks and the reduction of the time needed by physicians to provide consultation to patients. They will then spend their Year 4 back at the Department of Physics, possibly utilizing the acquired AI knowledge to further improve the sensitivity of the Higgs-pair production analysis toward the final publication foreseen for 2025-2026.