Development and use of new and derived data to facilitate the optimisation and evaluation of breast screening AI tools

Artificial Intelligence (AI) tools for diagnosis of cancer on breast screening images are becoming available. It is vital that these tools are able to be applied and effective in any setting across the UK and must be adaptable to varying populations, manufacturers, image processing and varying image acquisition techniques. It is therefore important that appropriately diverse and representative training and validation datasets are available. This can be achieved through the large-scale collection of de-identified data from large numbers of sites, and this is what the OPTIMAM Mammographic Image Database (OMI-DB) (containing 3.5 million images from over 200,000 women) is attempting to achieve. However, there are additional techniques that can be applied to supplement this dataset, enriching it so that it can be used to improve and optimise AI tools.

This PhD project will use techniques to alter an unprocessed image to make it appear as if it was captured with lower dose or different beam quality or x-ray detector, or processed using different manufactures image processing software to generate additional data to supplement OMI-DB. It will also develop image processing software to understand the effect of this on the images that are used for AI tool development. This project will involve the use of this data to investigate and evaluate how varying factors in the images may confound or affect an AI tools performance. This will involve retraining in-house algorithms and subsequent validation to evaluate the resulting effects that varying the training and validation datasets have.  Is it anticipated that in addition to the use of in-house or open-source algorithms the project will involve utilising the groups' existing collaboration with Google Health to investigate these same factors in the Google Health Breast AI tool.

This project will develop physics-based methodology to derive new data to assess AI based tools for medical imaging. This requires the development of new knowledge and fundamental research in a way which is well-suited to a PhD project. The project brings together metrological analysis and data science skills with medical imaging, and involves working with specialists from several different disciplines, preparing the candidate for a career in either medical research or data-intensive metrology.

The project is a collaboration between the University of Surrey (Radiation and Medical Physics Group), the National Physical Laboratory (Data Science and Medical Physics) and the Royal Surrey NHS Foundation Trust (Scientific Computing). The student will be predominately based at the University of Surrey and RSNFT with possibility of placements at NPL for up to 3 months a year.

UK applicants who hold a First or 2:1 UK honours degree in a relevant subject area, or a 2:2 alongside a good masters degree (a distinction is usually required).

English language requirements: If English is not your first language, you will be required to have an IELTS Academic of 6.5 or above (or equivalent), with no sub-test score below 6.

Radiation and Medical Physics