STUDENTSHIP IN DEEP GENERATIVE MODELLING OF BRAIN COMPUTED TOMOGRAPHY FOR UNIVERSAL CLINICAL TRIAGE
The speed and facility of computed tomography of the brain has preserved its position as the first investigation of choice across a wide range of clinical scenarios in neurology, neurosurgery, and acute medicine. This magnifies the clinical gain from enhancing any decision-making based on it, and expands the breadth of available data on which creating complex, high-dimensional models critically depends. Though tools for automated analysis within niche clinical contexts are emerging, no robust framework for universal characterization of the state of brain from computed tomography exists. The task is complicated by the extreme diversity of anatomical appearances and their interaction with instrumental factors, necessitating more complex architectures than have hitherto been computationally feasible. Deep learning offers possible solutions, but is vulnerable to poor generalization even with large-scale datasets. Our hybrid approach seeks to combine the merits of each with the demerits of neither, building on our success with 3D variational autoencoder-based architectures applied to registered multimodal clinical MR imaging.
This studentship will pursue such a hybrid approach to develop high-dimensional generative models of computed tomography brain appearances on which decision-making tools can be founded with enhanced generalizability. The focus will be on enabling instant, universal clinical triage into major management pathways, enabling rapid optimisation of care with immediate impact on patient outcomes. Development will draw on parallel advances in generative modelling of magnetic resonance imaging within our group, exploiting abstract correspondences between modalities to boost performance. A real-world implementation at our partner hospitals, within the context of operational use, will be delivered in the final third of the project.
We particularly welcome applications from individuals with a strong background in computing, machine learning, mathematical, or physical sciences, and highly motivated to pursue research with near-term, real-world translational objectives.
The studentship will be based at the UCL Wellcome Trust Centre for Neuroimaging and the High-Dimensional Neurology Group at the UCL Institute of Neurology, with funding from the UCLH NIHR Biomedical Research Centre and the Engineering and Physical Sciences Research Council.
The successful candidate will join a multi-disciplinary team applying novel modelling approaches to very large-scale clinical data for real-world, translational applications. The project is aligned with UCLH’s Research Hospital Initiative, which seeks to accelerate the introduction of artificial intelligence within the hospital for clinical, operational, and scientific benefit. It will draw on state-of the-art expertise in classical generative modelling of brain imaging (John Ashburner’s group), and deep generative modelling of clinical imaging (Parashkev Nachev’s group). It will be enabled by large-scale clinical data from UCH and NHNN traversing an arguably unique breadth of brain imaging appearances, and will be powered by a 1.5 petaFLOP, hospital-embedded GPU-based computational infrastructure.
The project represents an ideal opportunity for a talented and highly-motivated individual to develop a breadth of modelling skills, working on unique datasets, assisted by high-performance hardware, with real-world, near-term, practical impact on healthcare.
The candidate must be UK passport holders, or be an EU citizen who has been in the UK for 3 consecutive years at the time of application. The required qualifications can be found here.
How to Apply
Interested applicants are invited to email Professor John Ashburner (email@example.com) or Dr Parashkev Nachev (firstname.lastname@example.org) to discuss the applications informally. Please include, in the first correspondence, a copy of the CV and a one-page letter of motivation explaining your interest and fit to the studentship.