PhD Studentship: R2T2-Med: Resource-Aware AI for Scalable and Deployable Medical Image Analysis

Medical imaging generates very large and rapidly growing volumes of data. Although AI can support automated image analysis, many current models require substantial computing power, memory and processing time. This limits population-scale research, increases deployment costs and restricts adoption in healthcare settings with limited hardware or connectivity. There is therefore a need for medical AI that is not only accurate, but also efficient, reliable and practical to deploy. Research aim and details: This project aims to develop a resource-aware AI framework for both 2D and 3D medical imaging. The framework will adapt its computational requirements to the complexity and uncertainty of each case, the clinical risk and the available hardware.

The project will investigate where unnecessary computation occurs within medical imaging models and how it can be reduced without losing clinically important information. It will then develop a flexible model family that can be trained once and configured for different resource budgets. Safety will be embedded Engineering and Physical Sciences Research Council Doctoral Landscape Award throughout the research by assessing whether efficiency measures affect particular anatomies, clinical sites or patient groups disproportionately. It will be validated in two complementary settings: population-scale abdominal MRI analysis using UK Biobank and near-patient dermatology imaging using public and established clinical data pathways.

The research will also examine whether improved efficiency can be achieved without reducing performance or disproportionately affecting particular clinical sites or patient groups. Expected outcomes: The project will deliver an adaptive and deployable medical imaging AI framework, reproducible efficiency benchmarks, reusable software and evidence from both large-scale and resource-constrained settings. It targets substantial reductions in processing time and memory while maintaining accuracy, reliability and subgroup safety. The resulting technology will reduce computational and hardware requirements, enabling more scalable, sustainable and equitable use of medical imaging AI across research, NHS and commercial healthcare environments.

Funded by: EPSRC Engineering and Physical Sciences Research Council