FindAPhD Weekly PhD Newsletter | JOIN NOW FindAPhD Weekly PhD Newsletter | JOIN NOW

PhD in Physics and Astronomy - Synthetic retinal image data generation and analysis with machine learning


   College of Science and Engineering

This project is no longer listed on FindAPhD.com and may not be available.

Click here to search FindAPhD.com for PhD studentship opportunities
  Prof Ik Siong Heng, Dr Peter Wakeford, Dr Christopher Messenger, Dr J Van Hemert  No more applications being accepted  Competition Funded PhD Project (UK Students Only)

About the Project

Medical imaging has adopted deep learning techniques for numerous applications, from image pre-processing to automated disease diagnosis. Deep learning models are created and optimised using data and in general, the more data, the higher the performance attained by these models. However, data in medical imaging is expensive to obtain – images are typically collected via lengthy and expensive clinical studies, and the additional requirement to have experts annotate the images as input to the deep learning process ramps up the cost. Image collection and preparation is therefore a significant and substantial cost—in terms of both time and money—in the development of image analysis algorithms for medical diagnostic purposes.

This industry-partnered project aims to build and assess tools that generate data key to the development of retinal image analysis algorithms. One method to increase the amount of data for model building is to create synthetic data with generative deep learning models, such as generative adversarial networks (GANs) and flow-based generative models. This project will explore the training of generative models to create not only synthetic retinal image data but also corresponding expert-level annotations for each synthetic image. The image and annotation pairs will be used to train secondary deep learning models. The dependency of secondary model performance on the amount of synthetic data generated from different-sized subsets of real data will be measured. The student will determine what factors limit the performance gains achieved by using synthetic data augmentation, and whether these limits can be overcome.

Wider applications of deep learning in retinal imaging will also be explored, for example the transfer of models between imaging domains, architecture pruning and optimisation, image denoising and colour-balancing. Upon completion of this degree, the student will be an expert in applied deep learning techniques in the field of retinal image analysis—a mature but rapidly evolving field with areas that include telemedicine, automated disease diagnosis and grading, and other clinician-assisting technologies.

This project will be undertaken in collaboration with Optos plc (a Nikon company). Optos is a global leader in the design and manufacture of medical retinal imaging systems, dedicated to the mission of “Saving sight, saving lives”. Optos and the University of Glasgow’s School of Physics and Astronomy have a history of successful collaborations, which includes three graduated, co-supervised PhD students. Optos co-supervises about five PhD students at any time, which enables you to interact with students in various stages of their project. With  the University of Glasgow’s Institute for Gravitational Research experience with GANs and other machine learning approaches in gravitational wave data analysis, they will be the technological lead, and Optos will bring domain-specific knowledge and clinical data to the project.

Academic criteria

We expect candidates to hold a 2:1 degree or higher in a related STEM subject, although applications will be reviewed on a case-by-case basis.

How to Apply

Please refer to the following website for details on how to apply:

http://www.gla.ac.uk/research/opportunities/howtoapplyforaresearchdegree/.


Funding Notes

Funding is available to cover tuition fees for UK applicants or EU applicants with settled status in the UK for 3.5 years as well as paying a stipend at the Research Council rate (estimated £15,840 for Session 2022-23).
PhD saved successfully
View saved PhDs