Uncertainty quantification in computer vision and natural language processing for medical imaging

This project aims to develop novel uncertainty methodologies and a software toolkit with uncertainty awareness for radiograph-based disease detection. The proposed models are expected to differentiate themselves with the uncertainty quantification algorithms to address data annotation. This will be achieved through three concrete and actionable research tasks in the duration of the studentship: 1) data uncertainty (from images, texts) integration for medical imaging; 2) model uncertainty quantification; and 3) clinician-in-the-loop AI (incorporating knowledge into the AI). Thus, in identified high uncertainty cases, human validation, intervention, and more extensive tests can be carried out to avoid potential error. The resulted open-source software toolkit will be validated through an AI-assisted radiograph-based dental disease detection application and is transferable across a wide range of diagnostic radiology applications.

The application is rolling-based with no fixed submission deadline until the position is filled. Early applications are strongly encouraged for early PhD start. The PhD student will be based at the Nature Inspired Computing and Engineering (NICE) research group in the Department of Compute Science at the University of Surrey. The student will also benefit from ample computing and research resources from Centre for Vision, Speech and Signal Processing (CVSSP) and the Surrey Institute for People-Centred AI.

More about the Dental Disease Detection project.

Later start dates may be possible: April 2024, July 2024

Open to any UK or international candidates.

Applicants are expected to hold a Bachelor’s degree or above in Computer Science, Electrical Engineering, Statistics, Mathematics, Physics or similar (a First Class or good Upper Second Class Honours degree, or the equivalent from an overseas university). Past research experience on machine learning, computer vision, natural language processing, or Bayesian inference preferred.