PhD Studentship - Generative AI and Active Learning for Foundation Models Applied to Automated Segmentation of Multi-modal Images

Research theme: AI and medical imaging

How to apply: uom.link/pgr-apply [uom.link]

This project is fully funded for home and overseas students (funding is provided by the Doctoral Training Program plus AstraZeneca funding). Tuition fees will be paid and you will receive a tax free allowance (depending on circumstances) set at the UKRI rate (£19,237 for 2024/25). The duration of the PhD project is 3.5 years and the start date is January 2025.

This proposal focuses on automatically segmenting diverse medical imaging modalities such as MRI, CT, and pathology through the combined use of Foundation models, Active Learning and Generative AI (Artificial Intelligence). Automatic segmentation helps to reduce the workload on domain experts, including radiologists, in vivo specialists, and pathologists. Training reliable models requires enormous amounts of domain experts’ annotations, which are costly and time-consuming.

Foundation models are primarily trained on natural images, and the project will look at using these models for multi-modal medical images by first enhancing them using active learning and generative AI. We hypothesise that foundation models learn underlying global structures, with transfer learning, using domain-specific images, helping to learn local structures. For example, human MRI and CT data are more widely available than rat and mouse data. Developing methods to transfer learning from human data to animal data would optimise animal use and may bridge the gap between preclinical and clinical research.

However, Foundation models are primarily trained on massive datasets, which are not always accessible. We will use a combination of active learning and generative AI to provide domain-specific images to fine-tune the pre-trained foundation models. Active learning will help determine which images need expert labelling, and generative AI will help the expert better enhance the resolution of specific images.

Eligibility

• Applicants should have, or expect to achieve, at least a 2.1 honours degree or a master’s in a relevant science or engineering-related discipline.
• Ideally, applicants should have research experience evidenced by publications in journals or conferences.

Please contact Dr. Mauricio A Álvarez before you apply: mauricio.alvarezlopez@manchester.ac.uk

Advert information