Supervisor: Dr Martin Benning
Project description:
Medical imaging is the umbrella term for imaging patients with non-invasive techniques like ultrasound, magnetic resonance imaging or positron emission tomography. Each of these techniques requires the solution of an inverse problem to acquire data from indirect measurements, followed by a subsequent analysis of the data, e.g., to support a physician’s diagnosis of a heart condition. However, the inversion process is usually unstable with respect to measurement errors, which could potentially lead to wrong diagnoses, and it usually does neglect variability in patient data such as sex, ethnicity or age.In this project, the goal is to utilise generative modelling to perform the task inverse to the process described above: to generate artificial imaging data from individuals with different physical attributes that is coherent with real imaging data. In other words, the goal is to develop digital imaging twins that consider patient variability. Subsequently, this research project aims at the robust inversion of this artificial imaging data with convergent regularisation methods. Applying this robust inversion to real data would aid the diagnosis of medical conditions from imaging data without neglecting patient variability.To achieve these goals, the student will develop neural network-based methods to generate artificial imaging data. To enable a robust inversion of these samples, they will develop networks based on convergent regularisations. Regularisations are families of continuous operators that approximate unstable inverses of models in robust fashion. These regularisation-based networks will be studied mathematically, implemented computationally and be applied to select datasets in medical imaging.The project aligns with the themes of AI & Data Modelling and Fundamental Discovery Science and impacts medical imaging applications and processes that utilise medical imaging (e.g., subsequent diagnosis of heart conditions) alike. It is timely because the negligence of differences in sex and ethnicity pose long-standing challenges in the diagnosis of various diseases.
Further information:
How to apply
Entry requirements
Fees and funding